Downloads · 30 days
20
35% of all-time downloads
budget-internalization-iclr2027
budget-internalization-iclr2027/qwen3.5-4b-8k-drgrpo-equalminnow-s300
qwen3.5-4b-8k-drgrpo-equalminnow-s300 is a image-text-to-text model from budget-internalization-iclr2027. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as apache-2.0.
RL-finetuned Qwen/Qwen3.5-4B trained for math reasoning under a 8,192-token generation budget. Released as part of an anonymous ICLR 2027 submission.
Downloads · 30 days
20
35% of all-time downloads
All-time downloads
57
Public
Parameters
4.5B
9.1 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors9.1 GB · 100%
From the Hugging Face model README
equalminnowRL-finetuned Qwen/Qwen3.5-4B trained for math reasoning under a 8,192-token generation budget. Released as part of an anonymous ICLR 2027 submission.
Run ID (petname): equalminnow · checkpoint step 300
| Base model | Qwen/Qwen3.5-4B |
| Algorithm | Dr. GRPO: GRPO without reward std-normalization and with a constant length normalizer. |
Generation budget (max_new_tokens) | 8,192 |
| Data | DeepScaleR (math), 3 epochs max |
| Batch | 32 prompts × 8 rollouts per step |
| Optimizer | Adam, cosine LR schedule, peak LR 5e-7, 10 warmup steps |
| Steps | 300 |
| Reward | binary answer correctness (\boxed{} extraction) |
| Weights dtype | BF16 |
Training prompt (user turn, rendered with the base model's chat template):
Think step-by-step to solve the following problem. Output your answer inside of \\boxed{} tags.:
{problem}
Let's think step-by-step
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "budget-internalization-iclr2027/qwen3.5-4b-8k-drgrpo-equalminnow-s300"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="auto", device_map="auto")
With vLLM: vllm serve budget-internalization-iclr2027/qwen3.5-4b-8k-drgrpo-equalminnow-s300
Inherits the license of the base model (Qwen/Qwen3.5-4B).