dsouzaJithesh commited on
Commit
265eb75
·
verified ·
1 Parent(s): 63a568a

Upload CrossCoder checkpoint version 0

Browse files
checkpoints/crosscoder_0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61174f62e6112f564f945f9fd38459d353b8e540370c6f3e00ca932b38ad62dd
3
+ size 268511695
checkpoints/crosscoder_0_cfg.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "seed": 49,
3
+ "device": "cuda:0",
4
+ "wandb_project": "CrossCoder_Model_Diffing",
5
+ "wandb_entity": "jithesh",
6
+ "wandb_run_name": null,
7
+ "checkpoint_path": "./checkpoints/",
8
+ "data_cache_dir": "./cache/",
9
+ "data_dir": "./data/",
10
+ "hook_point": "blocks.14.hook_resid_pre",
11
+ "dict_size": 16384,
12
+ "enc_dtype": "fp32",
13
+ "dec_init_norm": 0.08,
14
+ "sparsity_type": "batch_top_k",
15
+ "k_sparsity": 100,
16
+ "l1_coeff": 0.0,
17
+ "l1_warmup_fraction": 0.0,
18
+ "lr": 5e-05,
19
+ "beta1": 0.9,
20
+ "beta2": 0.999,
21
+ "lr_warmup_steps": 0,
22
+ "lr_decay_start_fraction": 0.8,
23
+ "num_tokens": 10000000,
24
+ "batch_size": 2048,
25
+ "seq_len": 1024,
26
+ "model_batch_size": 1,
27
+ "buffer_mult": 64,
28
+ "norm_batch_size": 16,
29
+ "log_every": 100,
30
+ "save_every": 5000,
31
+ "d_in": 1024
32
+ }