Skip to main content
huggingface
Projects
open-r1
Reports
Qwen2.5-Coder-3B SFT on Reasoning Datasets
Log in
Sign up
Share
Comment
Star
Qwen2.5-Coder-3B SFT on Reasoning Datasets
Lewis Tunstall
Created on February 16
|
Last edited on February 16
Comment
train/grad_norm
train/grad_norm
50
100
150
200
250
300
train/global_step
1
2
3
4
5
qwen-3b_s1k_v00.00
qwen-3b_s1k_v00.03
qwen-3b_s1k_v00.04
qwen-3b_s1k_v00.02
qwen-3b_s1k_v00.05
qwen-3b_s1k_v00.01
eval/loss
eval/loss
Select runs that logged eval/loss
to visualize data in this line chart.
train/learning_rate
train/learning_rate
50
100
150
200
250
300
train/global_step
0.000002
0.000004
0.000006
0.000008
0.00001
qwen-3b_s1k_v00.00
qwen-3b_s1k_v00.03
qwen-3b_s1k_v00.04
qwen-3b_s1k_v00.02
qwen-3b_s1k_v00.05
qwen-3b_s1k_v00.01
train/loss
train/loss
50
100
150
200
250
300
train/global_step
0.8
1
1.2
qwen-3b_s1k_v00.00
qwen-3b_s1k_v00.03
qwen-3b_s1k_v00.04
qwen-3b_s1k_v00.02
qwen-3b_s1k_v00.05
qwen-3b_s1k_v00.01
OpenThoughts (Instruct)
4
OpenThoughts (Base)
6
s1k (Instruct)
6
Add a comment