ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-GGUF
042
reasoning-gym-chain-sum-qwen3-0.6b-grpo — GGUF
GGUF quantizations of `ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model`, a GRPO (Group Relative Policy Optimization) reinforcement-learning fine-tune, converted with llama.cpp.
Available quantizations
Recommended default: Q4_K_M. For maximum fidelity use Q8_0.
Usage
llama.cpp
# One-shot
llama-cli -hf ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-GGUF --jinja -p "Your prompt here" -n 256
# Interactive chat
llama-cli -hf ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-GGUF --jinja -cnvOllama
ollama run hf.co/ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-GGUF:Q4_K_Mllama-cpp-python
from llama_cpp import Llama
llm = Llama.from_pretrained(
repo_id="ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-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_0when fidelity matters. - Inherits every limitation of the source checkpoint (`ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model`).
- Optimized for the
chain_sumarithmetic-reasoning task; capability on unrelated tasks is not guaranteed.
Citation
@misc{reasoning_gym_chain_sum_qwen3_0_6b_grpo_gguf,
author = {Ermia Azarkhalili},
title = {reasoning-gym-chain-sum-qwen3-0.6b-grpo — GGUF quantized},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/ermiaazarkhalili/reasoning-gym-chain-sum-qwen3-0-6b-grpo-model-GGUF}}
}