SFT-trained step-level critic (Qwen3-8B) for code agents on SWE-bench Verified. Companion to https://github.com/shubhamrgandhi/critic-training
Shubham Gandhi
shubhamrgandhi
·
AI & ML interests
AI4Code
Recent Activity
commentedon a paper 4 days ago
DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training commentedon a paper 4 days ago
DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training upvoted a paper 5 days ago
Steer, Don't Solve: Training Small Critic Models for Large Code AgentsOrganizations
models 14
shubhamrgandhi/Qwen3-4B-DPO-prm-pairs
Text Generation • 4B • Updated • 421
shubhamrgandhi/qwen3_5-4b-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only
Image-Text-to-Text • 5B • Updated • 4
shubhamrgandhi/qwen3_5-4b-full-sft-prm-r2egym-swebench-instructions-k5-cwm-plus-qwen
Image-Text-to-Text • 5B • Updated • 10
shubhamrgandhi/qwen3_5-9b-full-sft-prm-r2egym-swebench-instructions-k5-qwen-only
Image-Text-to-Text • 9B • Updated • 5 • 1
shubhamrgandhi/qwen3_5-9b-full-sft-prm-r2egym-swebench-instructions-k5-cwm-plus-qwen
Image-Text-to-Text • 9B • Updated • 5
shubhamrgandhi/qwen3-8b-full-sft-prm-r2egym-swebench-k5-opus-distill-32k-lr5e6-multiturn
Text Generation • 1B • Updated • 12
shubhamrgandhi/qwen3-8b-full-sft-prm-r2egym-k5-opus-distill-32k-lr5e6-multiturn
Text Generation • 1B • Updated • 8
shubhamrgandhi/qwen3-8b-full-sft-prm-r2egym-instructions-k10-opus-distill-32k-lr5e6-multiturn
Updated
shubhamrgandhi/qwen3-8b-full-sft-prm-r2egym-instructions-k10-opus-distill-32k-lr5e6-flattened
Text Generation • 1B • Updated • 8
shubhamrgandhi/qwen3-8b-full-sft-prm-opus-distill-32k-lr5e6-multiturn
Text Generation • 1B • Updated • 15
datasets 0
None public yet