Training checkpoint at step 22500
Browse files- config.json +39 -0
config.json
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"act_format": "linear",
|
| 3 |
+
"act_on": false,
|
| 4 |
+
"act_type": "associative",
|
| 5 |
+
"architectures": [
|
| 6 |
+
"InnerLoopARMTForCausalLM"
|
| 7 |
+
],
|
| 8 |
+
"attend_to_previous_input": false,
|
| 9 |
+
"attn_implementation": "flash_attention_2",
|
| 10 |
+
"base_model_config": null,
|
| 11 |
+
"base_model_name": "google/gemma-3-1b-it",
|
| 12 |
+
"constant_depth": false,
|
| 13 |
+
"correction": true,
|
| 14 |
+
"d_mem": 64,
|
| 15 |
+
"dtype": "bfloat16",
|
| 16 |
+
"freeze_mem": false,
|
| 17 |
+
"gating": false,
|
| 18 |
+
"layers_attr": "model.layers",
|
| 19 |
+
"max_hop": 4,
|
| 20 |
+
"memory_dtype": "bfloat16",
|
| 21 |
+
"model_dtype": "bfloat16",
|
| 22 |
+
"model_type": "armt",
|
| 23 |
+
"n_heads": 1,
|
| 24 |
+
"noisy_halting": false,
|
| 25 |
+
"num_mem_tokens": 32,
|
| 26 |
+
"segment_alignment": "left",
|
| 27 |
+
"segment_size": 1024,
|
| 28 |
+
"sliding_window_enabled": false,
|
| 29 |
+
"time_penalty": 0.0,
|
| 30 |
+
"transformers_version": "4.57.1",
|
| 31 |
+
"use_denom": true,
|
| 32 |
+
"use_sink": false,
|
| 33 |
+
"wrap_layers": null,
|
| 34 |
+
"wrap_pos": false,
|
| 35 |
+
"auto_map": {
|
| 36 |
+
"AutoConfig": "modeling_armt.ARMTConfig",
|
| 37 |
+
"AutoModelForCausalLM": "modeling_armt.InnerLoopARMTForCausalLM"
|
| 38 |
+
}
|
| 39 |
+
}
|