3l3ktr4/donorsim-qwen3-8b-modeAB-step51
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: 3l3ktr4/donorsim-qwen3-8b-modeAB-step51
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-10-22 15:23:20 +08:00
commit 39b8487f0a
12 changed files with 151916 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
README.md Normal file
View File

@@ -0,0 +1,16 @@
---
base_model: Qwen/Qwen3-8B
library_name: transformers
tags: [grpo, verl, donor-game, cooperation, reciprocity]
---
# donorsim-qwen3-8b-modeAB-step51
Qwen3-8B fine-tuned with GRPO (verl 0.7.1, LoRA r16/alpha32 merged into bf16 weights) on the
iterated Donor's Game, Stage 2 (group games, K in {2,4,6}, within-group partner rotation with
w = 1/(K-1)), mixed Mode A/B (50% of scenarios with real groupmate discussion).
True training step 51. Lineage: [donorsim-qwen3-8b-modeAB-step6](https://huggingface.co/3l3ktr4/donorsim-qwen3-8b-modeAB-step6)
(steps 1-6, fixed partner) -> steps 7-11 on 4 GPUs -> steps 12-51 on 8 GPUs (run
`qwen3_8b_recipT2_modeAB_from_step11_64steps_20260827_055004`, global_step_40).
Merged full weights, loadable directly with `transformers` or vLLM (no adapter needed).