jgcb00 commited on
Commit
58dadf4
·
verified ·
1 Parent(s): 6082ea1

Card: add 'Models used' section

Browse files

Names the generator, judge and tokenizer models (with links) for each pipeline stage.

Files changed (1) hide show
  1. README.md +10 -0
README.md CHANGED
@@ -295,6 +295,16 @@ text = f"Problem: {r['rewritten_problem']}\n\nSolution: {r['rewritten_solution']
295
  bad = d.filter(lambda r: r["judge_correctness"] == "no")
296
  ```
297
 
 
 
 
 
 
 
 
 
 
 
298
  ## How it was built
299
 
300
  **Stage 1 — seed generation.** 60,000,000 problems from 122 named generators across
 
295
  bad = d.filter(lambda r: r["judge_correctness"] == "no")
296
  ```
297
 
298
+ ## Models used
299
+
300
+ | Role | Model | Settings |
301
+ |---|---|---|
302
+ | Rewrite (stage 2) | [`RedHatAI/gemma-4-31B-it-FP8-dynamic`](https://huggingface.co/RedHatAI/gemma-4-31B-it-FP8-dynamic) (FP8 quantization of [`google/gemma-4-31B-it`](https://huggingface.co/google/gemma-4-31B-it)) | temperature 0.8, served with vLLM |
303
+ | Judge (stage 3) | [`RedHatAI/gemma-4-31B-it-FP8-dynamic`](https://huggingface.co/RedHatAI/gemma-4-31B-it-FP8-dynamic) (FP8 quantization of [`google/gemma-4-31B-it`](https://huggingface.co/google/gemma-4-31B-it)), the same model as the rewriter | temperature 0.1, strict-JSON rubric |
304
+ | Token counts (`token_count`) | Qwen2.5 tokenizer ([`Qwen/Qwen2.5-7B-Instruct`](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct)) | |
305
+
306
+ Seed problems (stage 1) come from deterministic code, not from a model.
307
+
308
  ## How it was built
309
 
310
  **Stage 1 — seed generation.** 60,000,000 problems from 122 named generators across