Modelpublic

ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF

sourceHugging Faceapache-2.0updated 3mo agoView on Hugging Face

0likes52downloads

Model Card

reasoning-gym-chain-sum-qwen3-1.7b-grpo — GGUF

GGUF quantizations of `ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo`, a GRPO (Group Relative Policy Optimization) reinforcement-learning fine-tune, converted with llama.cpp.

Field	Value
Source checkpoint	`ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo`
Base model	`Qwen/Qwen3-1.7B`
Training	GRPO (group-relative RL) on the `chain_sum` arithmetic-reasoning task
Dataset / environment	OpenEnv `reasoning_gym` — `chain_sum` (agentic env)
Quantization tool	llama.cpp `convert_hf_to_gguf.py` + `llama-quantize`

Available quantizations

File	Size	Notes
`reasoning-gym-chain-sum-qwen3-1-7b-grpo.q4_k_m.gguf`	1.11 GB (recommended)	4-bit K-quant medium; best size/quality balance
`reasoning-gym-chain-sum-qwen3-1-7b-grpo.q5_k_m.gguf`	1.26 GB (balanced)	5-bit K-quant medium; near-full quality
`reasoning-gym-chain-sum-qwen3-1-7b-grpo.q8_0.gguf`	1.83 GB (largest)	8-bit; closest to the source precision

Recommended default: Q4_K_M. For maximum fidelity use Q8_0.

Usage

llama.cpp

bash

# One-shot
llama-cli -hf ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF --jinja -p "Your prompt here" -n 256

# Interactive chat
llama-cli -hf ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF --jinja -cnv

Ollama

bash

ollama run hf.co/ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF:Q4_K_M

llama-cpp-python

python

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF",
    filename="*q4_k_m.gguf",
    n_ctx=4096,
)
out = llm.create_chat_completion(
    messages=[{"role": "user", "content": "Your prompt here"}],
    max_tokens=256,
)
print(out["choices"][0]["message"]["content"])

Intended use

Research and non-commercial experimentation. This model was RL-tuned on the chain_sum arithmetic-reasoning task; it is a reasoning demonstrator, not a general-purpose assistant. Verify outputs before any downstream use.

Limitations

—GGUF quantizations carry unavoidable quality loss relative to the source weights; prefer Q8_0 when fidelity matters.
—Inherits every limitation of the source checkpoint (`ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo`).
—Optimized for the chain_sum arithmetic-reasoning task; capability on unrelated tasks is not guaranteed.

Citation

bibtex

@misc{reasoning_gym_chain_sum_qwen3_1_7b_grpo_gguf,
  author       = {Ermia Azarkhalili},
  title        = {reasoning-gym-chain-sum-qwen3-1.7b-grpo — GGUF quantized},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-1-7b-grpo-GGUF}}
}