tinystories-1M-SAES / 3_cfg.json
ckkissane's picture
First model version
d8bc519
raw
history blame
481 Bytes
{"seed": 49, "batch_size": 4096, "buffer_mult": 384, "lr": 0.001, "num_tokens": 200000000, "l1_coeff": 0.068, "beta1": 0.9, "beta2": 0.99, "dict_mult": 4, "seq_len": 128, "enc_dtype": "fp32", "model_name": "roneneldan/TinyStories-1M", "site": "mlp_out", "layer": 0, "device": "cuda:0", "model_batch_size": 512, "buffer_size": 1572864, "buffer_batches": 12288, "act_name": "blocks.0.hook_mlp_out", "act_size": 64, "dict_size": 256, "name": "roneneldan/TinyStories-1M_0_256_mlp_out"}