Downloads · 30 days
9
20% of all-time downloads
gabriel-xiong/qwen3-8b-grpo-v5-epoch3
qwen3-8b-grpo-v5-epoch3 is a machine learning model from gabriel-xiong. 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.
GRPO-trained memory-update head for the LLM-as-RNN clinical pipeline. LoRA merged into base model for direct inference (no PEFT runtime needed).
Downloads · 30 days
9
20% of all-time downloads
All-time downloads
46
Public
Parameters
8.2B
16.4 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors16.4 GB · 100%
From the Hugging Face model README
GRPO-trained memory-update head for the LLM-as-RNN clinical pipeline. LoRA merged into base model for direct inference (no PEFT runtime needed).
Qwen/Qwen3-8Bphase1_v5judge_run01_Qwen3_8b_3epochglobal_step_5168ca31d# Run manifest — generated by run_phase1.sh
project_name : llmrnn_grpo
experiment_name : phase1_v5judge_run01_Qwen3_8b_3epoch
launched_at : 2026-06-05T22:14:58Z
hostname : c309-001.ls6.tacc.utexas.edu
# Code identity
git_sha : 68ca31d
git_status : dirty
launcher_script : /var/spool/slurmd/job3211128/slurm_script
# Judge / reward
rubric_yaml : training/configs/rubric_v5_perdim_only.yaml
rubric_yaml_sha1 : 70c9b726d056b1d2ee697de8448481c4589fad14
judge_model : OpenRubrics/RubricARM-8B-Judge
judge_endpoint : http://localhost:8001
# Data
train_parquet : /scratch/11566/gabriel_xiong/data/llm_as_rnn/train.parquet
val_parquet : /scratch/11566/gabriel_xiong/data/llm_as_rnn/train.parquet
# Policy + GRPO
base_model : Qwen/Qwen3-8B
lora_rank : 16
lora_alpha : 32
lr : 3e-6
kl_coef : 0.001
kl_loss_type : low_var_kl
G_rollouts : 8
train_batch : 16
ppo_mini_batch : 16
total_epochs : 3