Downloads · 30 days
14
33% of all-time downloads
Ram20307/qwen3-0.6b-gsm8k-arm-a-seed123
qwen3-0.6b-gsm8k-arm-a-seed123 is a machine learning model from Ram20307. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Part of the SLM Reasoning Research project — studying what kind of reasoning supervision content actually helps a small (0.6B) language model learn to reason.
Downloads · 30 days
14
33% of all-time downloads
All-time downloads
43
Public
Parameters
596M
2.4 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors2.4 GB · 100%
From the Hugging Face model README
Part of the SLM Reasoning Research project — studying what kind of reasoning supervision content actually helps a small (0.6B) language model learn to reason.
Arm: A
Full reasoning (everything, including reflection/verification) — teacher-distilled from GPT-OSS-20B.
GSM8K test accuracy: 60% (60/100, n=100 subset).
{
"arm": "A",
"seed": 123,
"data_file": "results/exp05_arms_acd/arm_a_full_n7435.jsonl",
"n_examples": 7435,
"model_name": "Qwen/Qwen3-0.6B-Base",
"epochs": 3,
"learning_rate": 2e-05,
"per_device_batch_size": 2,
"grad_accum_steps": 8,
"effective_batch_size": 16,
"max_length": 768,
"optimizer": "adamw_torch",
"precision": "fp32_master_amp_fp16",
"gpu_name": "NVIDIA RTX A5000 Laptop GPU",
"python_version": "3.10.12",
"platform": "Linux-6.8.0-138-generic-x86_64-with-glibc2.35",
"torch_version": "2.5.1+cu124",
"transformers_version": "5.3.0",
"cuda_version": "12.4",
"git_commit": null,
"start_time": "2026-09-09T09:49:03.919665+00:00",
"end_time": "2026-09-09T11:22:14.742424+00:00",
"duration_seconds": 5590.822759,
"final_train_loss": 0.42533337223914364,
"train_runtime_seconds": 5579.5586,
"train_samples_per_second": 3.998
}
See the project repo for the full methodology, literature grounding, and comparison against the other reasoning-format arms.