hellouniverse/MiMo-SFT2-minerva-responses

MiMo SFT2 — Minerva Math responses (128× sampling, temp 0.6)

Model responses generated with an SFT2 MiMo-7B model (tequila3009/sft2_mimo, weights under the sft2_mimo/ subdir) on the Minerva Math problem set.

Generation setup

Model	SFT2 MiMo-7B — tequila3009/sft2_mimo (sft2_mimo/)
Dataset	Minerva Math — 272 problems
Samples per problem	128
Total responses	34,816
Temperature	0.6
top_p	0.95
top_k	-1 (disabled)
max_tokens	16384
Engine	vLLM, TP=8 on 8× A100

Each problem is wrapped in a natural-language chain-of-thought prompt that asks the model to reason step by step and put the final answer in \boxed{...}.

File

• sft2_mimo_minerva_responses_t0.6_20260707_140420.json (~486 MB)

Structure

{
  "model": "sft2_mimo",
  "dataset": "minerva-math.parquet",
  "sampling_params": {"temperature": 0.6, "top_p": 0.95, "top_k": -1, "max_tokens": 16384},
  "total_questions": 272,
  "responses_per_question": 128,
  "generation_time": 24845.0,
  "results": [
    {"question": "<problem text>", "responses": ["<sample 1>", "... 128 samples ..."]}
  ]
}

results is aligned to the Minerva Math order; results[i]["responses"] holds the 128 samples for problem i.

Notes

• All 34,816 responses are non-empty (0 empty completions).
• Responses are raw model text (reasoning + \boxed{} answer), not verified for correctness.

Loading

import json
data = json.load(open("sft2_mimo_minerva_responses_t0.6_20260707_140420.json"))
print(data["total_questions"], data["responses_per_question"])   # 272 128
q0 = data["results"][0]
print(q0["question"])
print(q0["responses"][0])