Merge branch 'main' of http://szcjw:10001/cjw/amd-r9700-ai-toolboxes

# Conflicts: # README.md
x
2026-04-15 23:14:33 +08:00 · 2026-04-15 23:09:38 +08:00 · 2026-02-27 15:07:05 +00:00 · 2026-02-27 14:49:24 +00:00 · 2025-12-27 17:44:35 +00:00 · 2025-12-17 08:01:12 +00:00
839 changed files with 58563 additions and 1 deletions
@@ -0,0 +1,97 @@
+name: Build & Publish AMD R9700 Toolboxes
+
+on:
+  workflow_dispatch:
+    inputs:
+      backends:
+        description: >
+          Comma-separated backends to build (e.g. "rocm-7beta,rocm-7rc").
+          Use "all" to build everything.
+        required: false
+        default: all
+
+env:
+  DOCKERHUB_REPO: docker.io/kyuz0/amd-r9700-toolboxes
+  LOCAL_PREFIX: llama
+
+jobs:
+  # 1) Prepare a clean JSON array for the matrix
+  prepare:
+    runs-on: ubuntu-latest
+    outputs:
+      matrix_json: ${{ steps.mk.outputs.matrix_json }}
+    steps:
+      - id: mk
+        shell: bash
+        run: |
+          # Input from the Run workflow form
+          IN='${{ inputs.backends }}'
+
+          if [[ "$IN" == "all" || -z "$IN" ]]; then
+            JSON='["rocm-6.4.4","rocm-7.2","rocm7-nightlies","vulkan-amdvlk","vulkan-radv"]'
+          else
+            # Remove spaces and build JSON array from comma list
+            IN_CLEAN=$(echo "$IN" | tr -d '[:space:]')
+            JSON='["'${IN_CLEAN//,/\",\"}'"]'
+          fi
+
+          echo "matrix_json=${JSON}" >> "$GITHUB_OUTPUT"
+          echo "Using matrix: ${JSON}"
+
+  # 2) Build each backend in parallel using the prepared matrix
+  build-and-push:
+    needs: prepare
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        backend: ${{ fromJson(needs.prepare.outputs.matrix_json) }}
+
+    steps:
+      - name: Free up runner disk space
+        run: |
+          echo "Before cleanup:" && df -h /
+          sudo rm -rf \
+            /usr/share/dotnet \
+            /usr/local/lib/android \
+            /opt/ghc \
+            /opt/hostedtoolcache/CodeQL
+          docker system prune --all --force
+          docker builder prune --all --force
+          echo "After cleanup:" && df -h /
+
+      - name: Check out repository
+        uses: actions/checkout@v3
+
+      - name: Log in to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Set build timestamp
+        run: echo "BUILD_TS=$(date +%Y%m%dT%H%M%S)" >> $GITHUB_ENV
+
+      - name: Build & push ${{ matrix.backend }}
+        working-directory: toolboxes
+        shell: bash
+        run: |
+          set -euo pipefail
+          B="${{ matrix.backend }}"
+          DF="Dockerfile.$B"
+          NAME="${B}"
+          LI="${LOCAL_PREFIX}-${NAME}"
+          TAG="${NAME}_${BUILD_TS}"
+          IMM="${DOCKERHUB_REPO}:${TAG}"
+          CHN="${DOCKERHUB_REPO}:${NAME}"
+
+          echo "→ Building ${DF}"
+          docker build --no-cache -t "${LI}" -f "${DF}" .
+
+          echo "→ Tag & push immutable → ${IMM}"
+          docker tag "${LI}" "${IMM}"
+          docker push "${IMM}"
+
+          echo "→ Tag & push channel → ${CHN}"
+          docker tag "${IMM}" "${CHN}"
+          docker push "${CHN}"
@@ -0,0 +1,105 @@
+name: Poll llama.cpp & Trigger Build
+
+on:
+  schedule:
+    - cron: '0 */4 * * *'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  actions: write
+
+jobs:
+  poll-and-trigger:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - id: fetch
+        shell: bash
+        run: |
+          set -euo pipefail
+          REPO_URL="https://github.com/ggml-org/llama.cpp.git"
+          DEFAULT_REF=$(git ls-remote --symref "$REPO_URL" HEAD | awk '/^ref:/ {print $2}')
+          echo "📌 Default branch: ${DEFAULT_REF#refs/heads/}"
+          LATEST_SHA=$(git ls-remote "$REPO_URL" "$DEFAULT_REF" | cut -f1)
+          if [[ -z "$LATEST_SHA" ]]; then echo "❌ No SHA found"; exit 1; fi
+          echo "✅ Latest SHA: $LATEST_SHA"
+          echo "latest_sha=$LATEST_SHA" >> "$GITHUB_OUTPUT"
+
+      - id: previous
+        shell: bash
+        env:
+          GH_REPO: ${{ github.repository }}
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          set -euo pipefail
+          echo "🔎 Checking for prior artifact 'last-llama-sha'…"
+          ART_ID=$(curl -fsSL -H "Authorization: Bearer $GH_TOKEN" \
+            "https://api.github.com/repos/${GH_REPO}/actions/artifacts?per_page=100" \
+            | jq -r '.artifacts | map(select(.name=="last-llama-sha" and .expired==false)) | sort_by(.created_at) | reverse | .[0].id // empty')
+          if [[ -n "$ART_ID" ]]; then
+            echo "📦 Found artifact id: $ART_ID (downloading)"
+            curl -fsSL -H "Authorization: Bearer $GH_TOKEN" -L \
+              "https://api.github.com/repos/${GH_REPO}/actions/artifacts/${ART_ID}/zip" -o artifact.zip
+            unzip -l artifact.zip || true
+            unzip -p artifact.zip last_commit_sha > last_commit_sha || true
+          else
+            echo "ℹ️ No prior artifact found"
+          fi
+          PREV_SHA=""
+          if [[ -f last_commit_sha ]]; then PREV_SHA=$(cat last_commit_sha); fi
+          echo "🕓 Previous SHA: $PREV_SHA"
+          echo "previous_sha=$PREV_SHA" >> "$GITHUB_OUTPUT"
+
+      - id: compare
+        shell: bash
+        run: |
+          set -euo pipefail
+          echo "🧮 Comparing SHAs…"
+          echo "prev: ${{ steps.previous.outputs.previous_sha }}"
+          echo "curr: ${{ steps.fetch.outputs.latest_sha }}"
+          if [[ "${{ steps.fetch.outputs.latest_sha }}" != "${{ steps.previous.outputs.previous_sha }}" ]]; then
+            echo "🔁 New commit detected"
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "✅ No change"
+            echo "changed=false" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Trigger build_and_publish.yml on main
+        if: steps.compare.outputs.changed == 'true'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          WF="build_and_publish.yml"
+          REF="main"
+          echo "🚀 Dispatching $WF on $REF…"
+          CODE=$(curl -s -o /tmp/resp -w "%{http_code}" \
+                 -X POST \
+                 -H "Accept: application/vnd.github+json" \
+                 -H "Authorization: Bearer $GH_TOKEN" \
+                 -d "{\"ref\":\"$REF\",\"inputs\":{\"backends\":\"all\"}}" \
+                 "https://api.github.com/repos/${{ github.repository }}/actions/workflows/$WF/dispatches")
+          echo "HTTP $CODE"
+          if [[ "$CODE" != "204" ]]; then echo "Response:"; cat /tmp/resp; exit 1; fi
+          
+
+      - name: Save new SHA
+        if: steps.compare.outputs.changed == 'true'
+        shell: bash
+        run: |
+          set -euo pipefail
+          printf "%s" "${{ steps.fetch.outputs.latest_sha }}" > last_commit_sha
+          echo "💾 Saved $(wc -c < last_commit_sha) bytes to $(pwd)/last_commit_sha"
+          ls -la last_commit_sha
+
+      - name: Upload last-SHA artifact
+        if: steps.compare.outputs.changed == 'true'
+        uses: actions/upload-artifact@v4
+        with:
+          name: last-llama-sha
+          path: last_commit_sha
+          retention-days: 7
@@ -0,0 +1,89 @@
+name: Prune Old Toolbox Images
+
+on:
+  workflow_dispatch:
+    inputs:
+      backends:
+        description: Comma-separated backends to prune (e.g. "rocm-7beta,rocm-7rc") or "all"
+        default: all
+      keep:
+        description: Number of latest tags to keep
+        default: "3"
+  workflow_run:
+    workflows: ["Build & Publish AMD Strix Halo Toolboxes"]
+    types: [completed]   # runs after success/failure/cancel
+    branches: [main]     
+
+jobs:
+  prune:
+    runs-on: ubuntu-latest
+    env:
+      NS: kyuz0
+      REPO: docker.io/kyuz0/amd-r9700-toolboxes
+    steps:
+      - name: Install jq
+        run: sudo apt-get update && sudo apt-get install -y jq
+
+      - name: Login to Docker Hub API (JWT)
+        id: login
+        env:
+          DH_USER: ${{ secrets.DOCKERHUB_USERNAME }}
+          DH_PASS: ${{ secrets.DOCKERHUB_TOKEN }}
+        run: |
+          TOKEN=$(curl -s -H "Content-Type: application/json" \
+            -d "{\"username\":\"${DH_USER}\",\"password\":\"${DH_PASS}\"}" \
+            https://hub.docker.com/v2/users/login/ | jq -r .token)
+          if [[ -z "$TOKEN" || "$TOKEN" == "null" ]]; then
+            echo "Failed to get Docker Hub JWT"; exit 1
+          fi
+          echo "token=${TOKEN}" >> "$GITHUB_OUTPUT"
+
+      - name: Determine backend list
+        id: mk
+        shell: bash
+        run: |
+          IN='${{ github.event.inputs.backends }}'
+          if [[ "$IN" == "all" || -z "$IN" ]]; then
+            JSON='["rocm-6.4.4","rocm-7.2","rocm7-nightlies","vulkan-amdvlk","vulkan-radv"]'
+          else
+            IN_CLEAN=$(echo "$IN" | tr -d '[:space:]')
+            JSON='["'${IN_CLEAN//,/\",\"}'"]'
+          fi
+          echo "list=${JSON}" >> "$GITHUB_OUTPUT"
+
+      - name: Prune old tags
+        env:
+          TOKEN: ${{ steps.login.outputs.token }}
+          KEEP: ${{ github.event.inputs.keep }}
+        run: |
+          BACKENDS='${{ steps.mk.outputs.list }}'
+          mapfile -t ARR < <(jq -r '.[]' <<< "$BACKENDS")
+          base_url="https://hub.docker.com/v2/repositories/${NS}/${REPO}/tags"
+          auth_hdr="Authorization: JWT ${TOKEN}"
+
+          for B in "${ARR[@]}"; do
+            echo ""
+            echo "=== Backend: ${B} (keeping latest ${KEEP}) ==="
+            next="${base_url}?page_size=100&ordering=last_updated&name=${B}_"
+            tags=()
+            while [[ -n "$next" && "$next" != "null" ]]; do
+              resp=$(curl -s -H "$auth_hdr" "$next")
+              page_tags=($(jq -r '.results[].name' <<< "$resp" | grep -E "^${B}_" || true))
+              tags+=("${page_tags[@]}")
+              next=$(jq -r '.next' <<< "$resp")
+            done
+            total=${#tags[@]}
+            echo "Found ${total} immutable tag(s) for ${B}."
+            if (( total <= KEEP )); then
+              echo "Nothing to delete."
+              continue
+            fi
+            to_delete=("${tags[@]:KEEP}")
+            for t in "${to_delete[@]}"; do
+              echo "Deleting tag ${t}..."
+              curl -s -X DELETE -H "$auth_hdr" \
+                "https://hub.docker.com/v2/repositories/${NS}/${REPO}/tags/${t}/" \
+                -o /dev/null -w "%{http_code}\n" | grep -Eq "^(202|204)$" \
+                && echo "✔ Deleted" || echo "✖ Failed"
+            done
+          done
@@ -0,0 +1 @@
+__pycache__
@@ -0,0 +1,5 @@
+# 默认忽略的文件
+/shelf/
+/workspace.xml
+# 基于编辑器的 HTTP 客户端请求
+/httpRequests/
@@ -0,0 +1,12 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<module type="PYTHON_MODULE" version="4">
+  <component name="NewModuleRootManager">
+    <content url="file://$MODULE_DIR$" />
+    <orderEntry type="jdk" jdkName="Python 3.13 (base)" jdkType="Python SDK" />
+    <orderEntry type="sourceFolder" forTests="false" />
+  </component>
+  <component name="PyDocumentationSettings">
+    <option name="format" value="PLAIN" />
+    <option name="myDocStringFormat" value="Plain" />
+  </component>
+</module>
@@ -0,0 +1,14 @@
+<component name="InspectionProjectProfileManager">
+  <profile version="1.0">
+    <option name="myName" value="Project Default" />
+    <inspection_tool class="PyPackageRequirementsInspection" enabled="true" level="WARNING" enabled_by_default="true">
+      <option name="ignoredPackages">
+        <list>
+          <option value="python-jose" />
+          <option value="passlib" />
+          <option value="aiosqlite" />
+        </list>
+      </option>
+    </inspection_tool>
+  </profile>
+</component>
@@ -0,0 +1,6 @@
+<component name="InspectionProjectProfileManager">
+  <settings>
+    <option name="USE_PROJECT_PROFILE" value="false" />
+    <version value="1.0" />
+  </settings>
+</component>
@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.13 (base)" project-jdk-type="Python SDK" />
+</project>
@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="ProjectModuleManager">
+    <modules>
+      <module fileurl="file://$PROJECT_DIR$/.idea/amd-r9700-ai-toolboxes.iml" filepath="$PROJECT_DIR$/.idea/amd-r9700-ai-toolboxes.iml" />
+    </modules>
+  </component>
+</project>
@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="VcsDirectoryMappings">
+    <mapping directory="" vcs="Git" />
+  </component>
+</project>
@@ -1,2 +1,148 @@
-# amd-r9700-ai-toolboxes
+# AMD R9700 Llama.cpp Toolboxes

+This project provides pre-built containers (“toolboxes”) for running LLMs on **AMD Radeon AI PRO R9700** GPUs (gfx1201). It uses `toolbox` (standard on Fedora, available on Ubuntu, Arch, etc.) to run `llama.cpp` with full GPU acceleration (Vulkan or ROCm) without messing up your host system.
+
+## Watch the YouTube Video
+
+[![Watch the YouTube Video](https://img.youtube.com/vi/dgyqBUD71lg/maxresdefault.jpg)](https://youtu.be/dgyqBUD71lg) 
+
+## 🚀 Quick Start
+
+### 1. Create a Toolbox
+**Which backend to choose?**
+*   **Vulkan (RADV)**: Recommended for **stability**. It works reliably with almost all models.
+*   **ROCm**: Recommended for **maximum performance**.
+    *   *Note*: Multiple ROCm versions are available (e.g., 6.4.4, 7.1, 7.9). Performance can vary significantly depending on the model architecture (e.g., Llama vs. Qwen). **Check the [Benchmarks](https://kyuz0.github.io/amd-r9700-ai-toolboxes/)** to find the best version for your model.
+
+**Option A: Vulkan (RADV) [Recommended]**
+```bash
+toolbox create llama-vulkan-radv \
+  --image docker.io/kyuz0/amd-r9700-toolboxes:vulkan-radv \
+  -- --device /dev/dri --group-add video --security-opt seccomp=unconfined
+```
+
+**Option B: ROCm (7.2)**
+```bash
+toolbox create llama-rocm-7.2 \
+  --image docker.io/kyuz0/amd-r9700-toolboxes:rocm-7.2 \
+  -- --device /dev/dri --device /dev/kfd \
+  --group-add video --group-add render --group-add sudo --security-opt seccomp=unconfined
+```
+
+> **Ubuntu Users**: `toolbox` may have issues with GPU access. Use [Distrobox](https://github.com/89luca89/distrobox) instead. See [Detailed Guide](#ubuntu-users-distrobox) below.
+
+### 2. Enter the Toolbox
+```bash
+toolbox enter llama-vulkan-radv
+# or: toolbox enter llama-rocm-7.2
+```
+
+### 3. Download a Model
+
+**Option A: Manual Download (Recommended)**
+Use the `hf` tool to download the model GGUF files to a local directory.
+
+```bash
+# Download to models/qwen3-coder-30B-A3B/
+HF_HUB_ENABLE_HF_TRANSFER=1 hf download unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF \
+  BF16/Qwen3-Coder-30B-A3B-Instruct-BF16-00001-of-00002.gguf \
+  --local-dir .
+```
+
+**Multi-shard Models:**
+If a model is split into multiple files (e.g., `00001-of-00005.gguf`), you must download **all** shards to the same folder. The command above ensures all parts are downloaded.
+
+> [!NOTE]
+> The old `huggingface-cli` is deprecated. Use the modern `hf` tool (part of `huggingface_hub`).
+
+**Option B: Automatic Download (via llama.cpp)**
+`llama.cpp` can automatically download models from the Hugging Face Hub to its internal cache (`~/.cache/huggingface/hub`).
+
+```bash
+# Automatically download and run
+llama-cli -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF -hf-file BF16/Qwen3-Coder-30B-A3B-Instruct-BF16-00001-of-00002.gguf ...
+```
+*Note: We prefer Option A (dedicated folder) to keep things organized, but Option B is great for quick tests.*
+
+### 4. Run a Model
+
+> [!TIP]
+> You should **always** use `-fa 1` (Flash Attention). This significantly improves performance and memory utilization on the R9700.
+
+Use **`llama-cli`** for running models directly in your terminal—ideal for quick tests, benchmarking, or chatting without leaving the shell.
+
+Use **`llama-server`** to start an OpenAI-compatible API server. This allows you to connect third-party UIs (like Open WebUI), use the built-in web interface, or build your own applications using standard libraries.
+
+**Run it (CLI Chat):**
+```bash
+llama-cli -ngl 999 -fa 1 \
+  -m models/qwen3-coder-30B-A3B/BF16/Qwen3-Coder-30B-A3B-Instruct-BF16-00001-of-00002.gguf \
+  -p "Write a R9700 toolkit haiku."
+```
+
+**Or run as Server (API + Web UI):**
+```bash
+llama-server -m models/qwen3-coder-30B-A3B/BF16/Qwen3-Coder-30B-A3B-Instruct-BF16-00001-of-00002.gguf \
+  -c 8192 -ngl 999 -fa 1
+```
+
+---
+
+## 📖 Detailed Guide
+
+### Managing Toolboxes
+
+#### Ubuntu Users (Distrobox)
+If you are on Ubuntu, use Distrobox to ensure proper GPU access:
+```bash
+distrobox create -n llama-rocm-7.2 \
+  --image docker.io/kyuz0/amd-r9700-toolboxes:rocm-7.2 \
+  --additional-flags "--device /dev/kfd --device /dev/dri --group-add video --group-add render --security-opt seccomp=unconfined"
+distrobox enter llama-rocm-7.2
+```
+
+#### Updating Toolboxes
+To pull the latest images and recreate your toolboxes (useful when Llama.cpp updates):
+```bash
+# Refresh all toolboxes
+./refresh-toolboxes.sh all
+
+# Or refresh specific ones
+./refresh-toolboxes.sh llama-vulkan-radv llama-rocm-7.2
+```
+
+## 📦 Architecture & Containers
+
+### Backends
+*   **Vulkan**: Cross-platform, very stable.
+    *   **RADV (Mesa)**: Best compatibility.
+    *   **AMDVLK**: Official AMD driver. Faster in some cases but has a strict 2GB single buffer limit (some large models won't load).
+*   **ROCm**: AMD's compute stack (CUDA-like).
+
+### Supported Container Images
+Images are hosted on [Docker Hub](https://hub.docker.com/r/kyuz0/amd-r9700-toolboxes/tags) and automatically rebuilt on Llama.cpp updates.
+
+| Tag | Backend | Notes |
+| :--- | :--- | :--- |
+| `vulkan-radv` | Vulkan (Mesa RADV) | Most stable and compatible. Recommended for most users and all models. |
+| `vulkan-amdvlk` | Vulkan (AMDVLK) | Fastest backend—AMD open-source driver. ≤2 GiB single buffer allocation limit, some large models won't load. |
+| `rocm-6.4.4` | ROCm 6.4.4 (Fedora 43) | Latest stable 6.x build. Uses Fedora 43 packages with backported patch for kernel 6.18.4+ support. |
+| `rocm-7.2` | ROCm 7.2 | Latest stable 7.x build. Includes patch for kernel 6.18.4+ support. |
+| `rocm7-nightlies` | ROCm 7 Nightlies | Nightly build for ROCm 7. |
+
+## ⚡ Performance & Planning
+
+### Benchmarks
+Check the [Interactive Benchmark Viewer](https://kyuz0.github.io/amd-r9700-ai-toolboxes/) or [docs/benchmarks.md](docs/benchmarks.md) to see performance numbers.
+
+### VRAM Estimator
+Use the included script to estimate memory usage for models + context. This helps avoid OOM errors.
+```bash
+gguf-vram-estimator.py models/my-model.gguf --contexts 4096 32768
+```
+See [docs/vram-estimator.md](docs/vram-estimator.md) for more details.
+
+##  References
+
+*   [Llama.cpp GitHub Repository](https://github.com/ggerganov/llama.cpp)
+*   [AMD RDNA™ 4 Architecture](https://www.amd.com/en/products/graphics/rdna-architecture.html)
@@ -0,0 +1,300 @@
+#!/usr/bin/env python3
+import re, glob, os, json, time
+from pathlib import Path
+
+RESULT_SOURCES = [
+    ("results", False),       # regular single-node runs
+]
+OUT_JSON = "../docs/results.json"
+
+# --- Regexes ---------------------------------------------------------------
+
+# Table headers come in two shapes (with or without "fa" column)
+HEADER_RE = re.compile(r"^\|\s*model\s*\|", re.IGNORECASE)
+SEP_RE    = re.compile(r"^\|\s*-+")
+
+# Build line, e.g. "build: cd6983d5 (6119)"
+BUILD_RE  = re.compile(r"build:\s*([0-9a-f]{7,})\s*\((\d+)\)", re.IGNORECASE)
+
+# Error classifiers (same spirit as your table script)
+LOAD_ERR   = re.compile(r"failed to load model|Device memory allocation.*failed|⚠️\s*Fail", re.IGNORECASE)
+HANG_ERR   = re.compile(r"GPU Hang|HW Exception", re.IGNORECASE)
+GENERIC_ERR= re.compile(r"error:|exit \d+|runtime error|⚠️\s*Runtime Error", re.IGNORECASE)
+
+# Extract numeric ± numeric from the last column
+TS_RE      = re.compile(r"([\d.]+)\s*±\s*([\d.]+)")
+
+# Quantization from model name
+QUANT_RE = re.compile(r"(Q\d+_[A-Z0-9_]+|BF16|F16|F32|mxfp\d+)", re.IGNORECASE)
+
+PARAMS_RE = re.compile(r"([\d.,]+)\s*B", re.IGNORECASE)
+GIB_RE    = re.compile(r"([\d.,]+)\s*GiB", re.IGNORECASE)
+
+# "30B", "235B" from model name
+NAME_B_RE = re.compile(r"(\d+(?:\.\d+)?)B")
+
+# Shard suffix in filenames
+SHARD_RE = re.compile(r"-000\d+-of-000\d+", re.IGNORECASE)
+
+# Long-context suffix in filenames (e.g., __longctx32768)
+LONGCTX_RE = re.compile(r"longctx(\d+)", re.IGNORECASE)
+
+# --- Helpers ---------------------------------------------------------------
+
+ENV_CANON = {
+    "rocm7_1": "rocm7.1",
+    "rocm7_alpha": "rocm-7alpha",
+}
+
+def clean_model_name(raw):
+    base = SHARD_RE.sub("", raw)
+    return base
+
+def canonicalize_env(env):
+    if not env:
+        return env
+    for raw, canon in ENV_CANON.items():
+        prefix = f"{raw}-"
+        if env == raw:
+            return canon
+        if env.startswith(prefix):
+            return canon + env[len(raw):]
+    return env
+
+def parse_env_flags(basename):
+    """
+    pattern: <model>__<env>[__fa1][__hblt0][__longctx32768][__rpc][__single][__dual]
+    Returns (env, fa, context_tag, context_tokens, rpc_flag, gpu_config)
+    """
+    parts = basename.split("__")
+    if len(parts) < 2:
+        return None, False, "default", None, False, "single"
+
+    env = parts[1]
+    fa = False
+    context_tag = "default"
+    context_tokens = None
+    rpc_flag = False
+    gpu_config = "single"  # default to single if not specified
+
+    for raw_suffix in parts[2:]:
+        suffix = raw_suffix.lower()
+        if suffix == "fa1":
+            fa = True
+        elif suffix == "hblt0":
+            env = f"{env}-hblt0"
+        elif suffix.startswith("longctx"):
+            context_tag = suffix
+            m = LONGCTX_RE.search(suffix)
+            if m:
+                try:
+                    context_tokens = int(m.group(1))
+                except ValueError:
+                    context_tokens = None
+        elif suffix == "rpc":
+            rpc_flag = True
+        elif suffix == "single":
+            gpu_config = "single"
+        elif suffix == "dual":
+            gpu_config = "dual"
+
+    return env, fa, context_tag, context_tokens, rpc_flag, gpu_config
+
+def env_base_and_variant(env):
+    # e.g. "rocm6_4_2-rocwmma" -> ("rocm6_4_2", "rocwmma")
+    if "-" in env:
+        base, variant = env.split("-", 1)
+        return base, variant
+    return env, None
+
+def detect_error(text):
+    if LOAD_ERR.search(text):
+        return True, "load"
+    if HANG_ERR.search(text):
+        return True, "hang"
+    if GENERIC_ERR.search(text):
+        return True, "runtime"
+    return False, None
+
+def parse_table(text):
+    """
+    Returns list of rows parsed from the markdown-like table.
+    Each row is a dict of the parsed columns, normalized by header names.
+    Handles presence/absence of the 'fa' column.
+    """
+    lines = text.splitlines()
+    rows = []
+    header = None
+    col_idx = {}
+
+    for i, line in enumerate(lines):
+        if HEADER_RE.search(line):
+            # header line
+            header = [c.strip().lower() for c in line.strip().strip("|").split("|")]
+            # next line should be the separator; skip it
+            # build index map
+            for idx, name in enumerate(header):
+                col_idx[name] = idx
+            continue
+        if header and (SEP_RE.search(line) or not line.strip()):
+            # skip separators / blanks after header
+            continue
+        if header and line.startswith("|"):
+            parts = [c.strip() for c in line.strip().strip("|").split("|")]
+            # guard for short lines
+            if len(parts) < len(header):
+                continue
+            row = {}
+            for name, idx in col_idx.items():
+                row[name] = parts[idx]
+            rows.append(row)
+        # stop parsing block when a blank line after some rows appears
+        if header and line.strip() == "" and rows:
+            break
+
+    return rows
+
+def coerce_float(m, default=None):
+    try:
+        return float(m)
+    except:
+        return default
+
+def extract_quant(model_name):
+    m = QUANT_RE.search(model_name)
+    return (m.group(1).upper() if m else None)
+
+def b_from_name(model_name):
+    m = NAME_B_RE.search(model_name)
+    return coerce_float(m.group(1)) if m else None
+
+# --- Main scan -------------------------------------------------------------
+
+runs = []
+builds = set()
+envs  = set()
+
+for results_dir, is_rpc_source in RESULT_SOURCES:
+    glob_pattern = os.path.join(results_dir, "*.log")
+    for path in sorted(glob.glob(glob_pattern)):
+        base = os.path.basename(path).rsplit(".log", 1)[0]
+        if "__" not in base:
+            continue
+
+        model_raw, _rest = base.split("__", 1)
+        env, fa_from_name, context_tag, context_tokens, rpc_flag, gpu_config = parse_env_flags(base)
+        env = canonicalize_env(env)
+        if env:
+            envs.add(env)
+
+        model_clean = clean_model_name(model_raw)
+
+        with open(path, errors="ignore") as f:
+            text = f.read()
+
+        # build info (take the last match in file if many)
+        build_hash, build_num = None, None
+        for m in BUILD_RE.finditer(text):
+            build_hash, build_num = m.group(1), m.group(2)
+        if build_hash:
+            builds.add((build_hash, build_num))
+
+        # detect error (if there is no valid table rows)
+        table_rows = parse_table(text)
+
+        # If table rows exist, we’ll still mark errors only if no perf found
+        has_pp = any(r.get("test","").lower()=="pp512" for r in table_rows)
+        has_tg = any(r.get("test","").lower()=="tg128" for r in table_rows)
+        error, etype = (False, None)
+        if not (has_pp or has_tg):
+            error, etype = detect_error(text)
+
+        # Determine FA flag:
+        #   prefer explicit column "fa" if present, else fallback to filename "__fa1"
+        fa_in_table = None
+        for r in table_rows:
+            if "fa" in r:
+                try:
+                    fa_in_table = int(r["fa"]) == 1
+                except:
+                    fa_in_table = None
+                break
+        fa_enabled = fa_in_table if fa_in_table is not None else fa_from_name
+
+        # Normalize env base / variant (e.g., rocwmma)
+        env_base, env_variant = env_base_and_variant(env)
+
+        # Emit one run per row (pp512 / tg128)
+        for r in table_rows or [{}]:
+            test = r.get("test", "").lower() if table_rows else None
+            tps_mean, tps_std = None, None
+            if table_rows:
+                ts_field = r.get("t/s", "")
+                m = TS_RE.search(ts_field)
+                if m:
+                    tps_mean = coerce_float(m.group(1))
+                    tps_std  = coerce_float(m.group(2))
+
+            # parse numeric helpers from row (if present)
+            params_b = None
+            file_size_gib = None
+            if "params" in r:
+                pm = PARAMS_RE.search(r["params"])
+                if pm:
+                    params_b = coerce_float(pm.group(1).replace(",", ""))
+            if "size" in r:
+                sm = GIB_RE.search(r["size"])
+                if sm:
+                    file_size_gib = coerce_float(sm.group(1).replace(",", ""))
+
+            # quant from model name (unchanged)
+            quant = extract_quant(model_clean)
+
+            # name_params_b: prefer table value; else fall back to B in model name
+            name_params_b = params_b if params_b is not None else b_from_name(model_clean)
+
+            backend = r.get("backend")
+            ngl = r.get("ngl")
+            mmap = r.get("mmap")
+
+            run = {
+                "model": model_raw,
+                "model_clean": model_clean,
+                "env": env,
+                "env_base": env_base,
+                "env_variant": env_variant,         # e.g. "rocwmma"
+                "fa": bool(fa_enabled),
+                "context": context_tag or "default",
+                "context_tokens": context_tokens,
+                "test": test,                       # "pp512" | "tg128" | None (if error)
+                "tps_mean": tps_mean,
+                "tps_std": tps_std,
+                "error": bool(error),
+                "error_type": etype,                # "load" | "hang" | "runtime" | None
+                "backend": backend,
+                "ngl": (int(ngl) if (ngl and ngl.isdigit()) else None),
+                "mmap": (int(mmap) if (mmap and mmap.isdigit()) else None),
+                "params_b": params_b,               # from table, if available
+                "file_size_gib": file_size_gib,     # from table, if available
+                "name_params_b": name_params_b,     # parsed from model name (e.g., 30B -> 30.0)
+                "quant": quant,
+                "log": path,
+                "rpc": bool(is_rpc_source or rpc_flag),
+                "gpu_config": gpu_config,
+                "build": {"hash": build_hash, "number": build_num} if build_hash else None,
+            }
+            runs.append(run)
+
+# Meta
+meta = {
+    "generated_at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+    "os_kernel": "Fedora 42 — Linux 6.15.9-201.fc42.x86_64 (Sat Aug  2 11:37:34 UTC 2025)",
+    "llamacpp_builds": [{"hash": h, "number": n} for (h, n) in sorted(builds)],
+    "environments": sorted(envs),
+    "notes": "pp512 = prompt processing; tg128 = text generation; t/s = tokens/second",
+}
+
+out = {"meta": meta, "runs": runs}
+
+Path(OUT_JSON).write_text(json.dumps(out, indent=2))
+print(f"Wrote {OUT_JSON} with {len(runs)} rows.")
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        251.12 ± 0.62 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.53 ± 0.03 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        124.60 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.52 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7-nightly-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        246.03 ± 0.16 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.41 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        123.04 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.46 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7-nightly-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        239.85 ± 0.14 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.37 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         99.59 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.82 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7-nightly] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        240.74 ± 0.07 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.41 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        100.10 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.83 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7-nightly] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        235.80 ± 0.18 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.01 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         95.03 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7.9-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        233.71 ± 0.16 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.94 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         94.33 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7.9-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        235.23 ± 0.14 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.94 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         98.01 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.45 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7.9] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        237.08 ± 0.15 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.00 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         99.40 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.47 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm-7.9] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        235.42 ± 0.14 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.02 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         94.65 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm6_4_4-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        233.43 ± 0.03 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.94 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         94.89 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm6_4_4-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        238.15 ± 0.10 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.99 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        109.79 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.50 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm6_4_4] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        239.15 ± 0.10 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.03 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        109.76 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.48 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm6_4_4] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        236.00 ± 0.17 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.01 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         94.52 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm7.1.1-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        233.56 ± 0.08 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.93 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         94.48 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.09 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm7.1.1-rocwmma] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        235.60 ± 0.08 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          6.93 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         99.54 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.47 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm7.1.1] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           pp512 |        237.37 ± 0.19 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |  1 |           tg128 |          7.01 ± 0.02 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,11 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |         99.62 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |          6.45 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,9 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 2 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+  Device 1: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+main: error: failed to create context with model '/home/kyuz0/models/Q3_K_S/Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002.gguf'
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+✖ ! [rocm7.1.1] Devstral-2-123B-Instruct-2512-Q3_K_S-00001-of-00002__hblt0__fa1 __longctx32768 failed (exit 0)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Ryzen 9 9900X3D 12-Core Processor (AMD open-source driver) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 32768 | int dot: 1 | matrix cores: none
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |           pp512 |        133.82 ± 0.17 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |           tg128 |          7.55 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Ryzen 9 9900X3D 12-Core Processor (AMD open-source driver) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 32768 | int dot: 1 | matrix cores: none
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 | pp2048 @ d16384 |         44.89 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |   tg32 @ d16384 |          4.37 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (AMD open-source driver) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 32768 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Ryzen 9 9900X3D 12-Core Processor (AMD open-source driver) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 32768 | int dot: 1 | matrix cores: none
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 | pp2048 @ d32768 |         16.45 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |   tg32 @ d32768 |          3.13 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Ryzen 9 9900X3D 12-Core Processor (RADV RAPHAEL_MENDOCINO) (radv) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 65536 | int dot: 1 | matrix cores: none
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |           pp512 |        104.90 ± 0.13 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |           tg128 |          8.35 ± 0.01 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Ryzen 9 9900X3D 12-Core Processor (RADV RAPHAEL_MENDOCINO) (radv) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 65536 | int dot: 1 | matrix cores: none
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 | pp2048 @ d16384 |         74.48 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |   tg32 @ d16384 |          6.44 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,12 @@
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+WARNING: radv is not a conformant Vulkan implementation, testing use only.
+ggml_vulkan: Found 3 Vulkan devices:
+ggml_vulkan: 0 = AMD Ryzen 9 9900X3D 12-Core Processor (RADV RAPHAEL_MENDOCINO) (radv) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 32 | shared memory: 65536 | int dot: 1 | matrix cores: none
+ggml_vulkan: 1 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+ggml_vulkan: 2 = AMD Radeon AI PRO R9700 (RADV GFX1201) (radv) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 | pp2048 @ d32768 |         36.00 ± 0.00 |
+| llama ?B Q3_K - Small          |  50.63 GiB |   125.03 B | Vulkan     |  99 |  1 |   tg32 @ d32768 |          2.94 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        964.08 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         19.68 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        571.14 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         17.45 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |      2843.52 ± 16.68 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.18 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        305.90 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         19.69 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        252.57 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         17.46 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |        380.51 ± 0.29 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.17 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        757.63 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.22 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        449.05 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.64 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |      2767.12 ± 15.29 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.21 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        279.80 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.23 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        222.90 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.62 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |        380.81 ± 0.21 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.21 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        731.41 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.15 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        419.35 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.45 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |       2608.48 ± 2.57 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.28 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        271.63 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.15 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        213.61 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.44 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |        368.45 ± 0.14 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.29 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        761.85 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.30 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        447.32 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.69 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |       2644.29 ± 2.60 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.32 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        274.70 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.30 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        219.43 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.70 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon Graphics, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |        370.52 ± 0.20 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.32 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        719.86 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.22 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        421.46 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.48 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |       2405.48 ± 3.53 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.36 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        276.54 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.21 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        217.88 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.48 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |        378.57 ± 0.27 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.35 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d16384 |        828.88 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d16384 |         20.37 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 | pp2048 @ d32768 |        501.73 ± 0.00 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |     2048 |  1 |   tg32 @ d32768 |         18.74 ± 0.00 |
+
+build: 2aa45ef9e (7423)
@@ -0,0 +1,10 @@
+ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
+ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
+ggml_cuda_init: found 1 ROCm devices:
+  Device 0: AMD Radeon AI PRO R9700, gfx1201 (0x1201), VMM: no, Wave Size: 32
+| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           pp512 |       2468.22 ± 5.84 |
+| mistral3 14B BF16              |  25.16 GiB |    13.51 B | ROCm       |  99 |  1 |           tg128 |         22.39 ± 0.01 |
+
+build: 2aa45ef9e (7423)
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
cjw	eb0d2ecf6d	Merge branch 'main' of http://szcjw:10001/cjw/amd-r9700-ai-toolboxes # Conflicts: # README.md	2026-04-15 23:14:33 +08:00
cjw	fa6bc9a404	x	2026-04-15 23:09:38 +08:00
Donato Capitella	8ad62bcda8	updating bench script	2026-02-27 15:07:05 +00:00
Donato Capitella	7e8603f91e	chore: Remove deprecated ROCm and rocWMMA Dockerfiles and associated build scripts, updating CI/CD workflows and documentation.	2026-02-27 14:49:24 +00:00
Donato Capitella	2d90fc1d90	added Devstral-2 to benchmarks	2025-12-27 17:44:35 +00:00
Donato Capitella	2e8bef127f	benchmakr updates	2025-12-17 08:01:12 +00:00
Donato Capitella	29da348bb8	Added YouTube Video reference	2025-12-10 17:16:58 +00:00
Donato Capitella	99d8dbe914	more banchmarks	2025-12-07 16:57:56 +00:00
Donato Capitella	672487a72b	feat: update the benchmark runner script.	2025-12-07 11:11:56 +00:00
Donato Capitella	ffbc6d98ff	updated benchmarks	2025-12-07 10:27:33 +00:00
Donato Capitella	204493bd4c	feat: Add ROCm 7.1.1 MMF toolbox and new NVIDIA benchmark comparison tool with value scoring.	2025-12-07 09:33:47 +00:00
Donato Capitella	689102e651	updated refresh and benchmarks scripts for ROCm 7.1.1	2025-12-07 09:10:04 +00:00
Donato Capitella	c8802647b9	feat: Add ROCm-optimized matrix multiplication with ID support, a ROCm 7.1.1 Dockerfile,, and update CI workflow.	2025-12-07 08:59:32 +00:00
Donato Capitella	341f837e6e	bump to rocm7.1.1 and fix pruning action	2025-12-07 08:54:29 +00:00
Donato Capitella	2a783325f0	added ministral-3-8B	2025-12-06 18:13:25 +00:00
Donato Capitella	9fcc0eac1c	updated benchmarks	2025-12-06 11:03:03 +00:00
Donato Capitella	a179482018	docs: streamline README with a concise quick start guide and remove detailed usage details.	2025-12-06 10:29:56 +00:00
Donato Capitella	cef614d895	updated benchmarks	2025-12-06 09:24:07 +00:00
Donato Capitella	5fe1510981	update qwen3-next-benchmarks	2025-12-05 08:44:12 +00:00
Donato Capitella	3d9087c39b	updated benchmarks	2025-11-30 13:47:32 +00:00
Donato Capitella	301424c1f7	updated broken link	2025-11-30 12:01:37 +00:00
Donato Capitella	17dbe11ddb	updated README	2025-11-30 11:54:30 +00:00
Donato Capitella	84d8d7ce5a	updated benchmarks	2025-11-30 11:44:58 +00:00
Donato Capitella	1b8bfa7230	added initial benchmark results	2025-11-30 11:17:29 +00:00
Donato Capitella	649cf0ab45	remove longcontext. enable memory mapping and point to ~/models	2025-11-30 10:05:06 +00:00
Donato Capitella	3f17169d17	feat: calculate total size for multi-shard models in benchmarks	2025-11-30 09:58:08 +00:00
Donato Capitella	f79c5099e2	fix: Remove redundant `env` keyword when setting ROCm environment variables.	2025-11-30 09:44:12 +00:00
Donato Capitella	5dfbe9f52e	build: Remove static libraries, headers, and documentation from ROCm installation.	2025-11-30 09:06:55 +00:00
Donato Capitella	aab95c2149	chore: Remove temporary ROCm tarball and rocWMMA build directory after installation.	2025-11-30 08:49:26 +00:00
Donato Capitella	0235d32a8e	chore: Remove ROCm tarball after extraction and standardize Dockerfile indentation.	2025-11-30 08:49:11 +00:00
Donato Capitella	7b4006b1d4	updated benchmarks	2025-11-30 08:20:12 +00:00
Donato Capitella	82c6e04253	first commit	2025-11-30 08:06:04 +00:00