Hugging Face's logo

Datasets:
RL-Forgetting-Experiments-3
/
code-sft-baselines-20261027-artifacts

Tags:
rl-forgetting
filtered-sft
License:
Dataset card Files
xet
 Community
smilani commited on
Commit
d8ffc73
·
verified ·
1 Parent(s): 483af43

Publish launch-records/support

Browse files
Files changed (2) hide show
  1. runs/launch-records/llama3b-code-ordered.json +29 -0
  2. runs/launch-records/llama3b-code-shuffled.json +29 -0
runs/launch-records/llama3b-code-ordered.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": {
3
+ "name": "llama3b-code-ordered",
4
+ "model": "llama3b",
5
+ "dataset": "llama3b_code_sft_data",
6
+ "base": "Llama-3.2-3B-Instruct",
7
+ "order": "ordered",
8
+ "strategy": "none"
9
+ },
10
+ "source_revision": "ec5a459cf6def8d1cf420b10a6b9ac5359681d9f",
11
+ "checkpoint_fix": "recovery-20261020",
12
+ "environment": {
13
+ "MODEL_NAME": "llama3b",
14
+ "DOMAIN": "code",
15
+ "DATASET": "llama3b_code_sft_data",
16
+ "BASE_MODEL": "/mnt/mystorageoutput/projects/rl-forgetting/base_models/Llama-3.2-3B-Instruct",
17
+ "LRS": "1e-5",
18
+ "ORDERS": "ordered",
19
+ "REPLAY_STRATEGY": "none",
20
+ "REPLAY_LAMBDA": "0",
21
+ "NPROC": "4",
22
+ "MICRO_BATCH_SIZE": "1",
23
+ "EFFECTIVE_BATCH_SIZE": "256",
24
+ "NUM_EPOCHS": "1",
25
+ "SEED": "42",
26
+ "MAX_SEQ_LENGTH": "4096",
27
+ "SKIP_EVAL": "1"
28
+ }
29
+ }
runs/launch-records/llama3b-code-shuffled.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": {
3
+ "name": "llama3b-code-shuffled",
4
+ "model": "llama3b",
5
+ "dataset": "llama3b_code_sft_data",
6
+ "base": "Llama-3.2-3B-Instruct",
7
+ "order": "shuffled",
8
+ "strategy": "none"
9
+ },
10
+ "source_revision": "ec5a459cf6def8d1cf420b10a6b9ac5359681d9f",
11
+ "checkpoint_fix": "recovery-20261020",
12
+ "environment": {
13
+ "MODEL_NAME": "llama3b",
14
+ "DOMAIN": "code",
15
+ "DATASET": "llama3b_code_sft_data",
16
+ "BASE_MODEL": "/mnt/mystorageoutput/projects/rl-forgetting/base_models/Llama-3.2-3B-Instruct",
17
+ "LRS": "1e-5",
18
+ "ORDERS": "shuffled",
19
+ "REPLAY_STRATEGY": "none",
20
+ "REPLAY_LAMBDA": "0",
21
+ "NPROC": "4",
22
+ "MICRO_BATCH_SIZE": "1",
23
+ "EFFECTIVE_BATCH_SIZE": "256",
24
+ "NUM_EPOCHS": "1",
25
+ "SEED": "42",
26
+ "MAX_SEQ_LENGTH": "4096",
27
+ "SKIP_EVAL": "1"
28
+ }
29
+ }