Repository navigation
README refresh: 50+ tok/s headline, recipe, new charts, project-scale card - #3
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c86bf48560
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| | Step | What it buys | Evidence | | ||
| |---|---|---| | ||
| | **1. DFlash2 draft model**, 7 tokens per round | Decode **16 -> 53 tok/s** mean (47-58), coding still 7/7 correct | [tune.txt](data/raw/tune.txt), [compare-f16.txt](data/raw/compare-f16.txt) | |
There was a problem hiding this comment.
Avoid presenting unlike decode workloads as an A/B
The 16 → 53 tok/s arrow implies a measured DFlash2 speedup on the same workload, but the new CSV shows that 53 is the mean of seven server-side coding requests while 16.05 comes from llama-bench tg32 at empty context. Because speculative-decoding acceptance and throughput depend on the generated workload, these figures cannot establish that before/after gain; either rerun the seven tasks without the drafter or label the values as separate measurements without the arrow.
Useful? React with 👍 / 👎.
| fig.text(0.04, 0.93, f"Every coding task decodes at 47-58 tok/s; {sum(v >= 50 for v in drafted)} of {len(drafted)} clear 50", | ||
| color=t["ink"], fontsize=16.5, fontweight="bold") |
There was a problem hiding this comment.
Derive the decode headline from the input rows
When data/decode-per-request.csv is updated with another task or new measurements, the bars and mean will regenerate from the CSV but this headline will continue to claim the fixed range 47-58. That silently produces a self-contradictory published chart despite running the documented generator; compute the displayed minimum and maximum from drafted just as the mean is computed.
Useful? React with 👍 / 👎.
|
|
||
| ```powershell | ||
| $env:GGML_VK_FORCE_MMVQ = "1" | ||
| llama-server -m Qwen3.8-27B-UD-Q4_K_M.gguf -ngl 99 -fa on -c 65536 -ctk f16 -ctv f16 ` |
There was a problem hiding this comment.
Use the paths created by the documented setup
For a user following the linked Quick start and docs/SETUP.md, this copy-paste command cannot start: the build writes the executable to C:\AI\llama.cpp\vulkan-intelfa\llama-server.exe, while the model and drafter are placed under C:\AI\models\qwen3.8-27b\; neither is added to PATH or copied into the repository working directory. Invoke the built executable and pass the documented absolute model paths so the advertised recipe works from the location where users run the README commands.
Useful? React with 👍 / 👎.
Front-page refresh to make the repo easier to find and act on for Intel Arc owners. No new measurements: every figure comes from existing logs.
Generated with Claude (Cowork).