checkpoint step 10000
Browse files
checkpoints/step-00010000/config.json
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"vocab_size": 256,
|
| 3 |
+
"sequence_length": 1024,
|
| 4 |
+
"local_dimension": 384,
|
| 5 |
+
"local_layers": 4,
|
| 6 |
+
"local_heads": 6,
|
| 7 |
+
"local_intermediate_size": 1536,
|
| 8 |
+
"dimension": 1024,
|
| 9 |
+
"heads": 16,
|
| 10 |
+
"intermediate_size": 12032,
|
| 11 |
+
"route_paths": 3,
|
| 12 |
+
"active_layers": 8,
|
| 13 |
+
"max_patches": 192,
|
| 14 |
+
"patch_rate": 0.125,
|
| 15 |
+
"entropy_threshold": 4.125,
|
| 16 |
+
"rope_theta": 10000.0,
|
| 17 |
+
"patch_aux_weight": 0.25,
|
| 18 |
+
"model_type": "cascade_byte_lm",
|
| 19 |
+
"architectures": [
|
| 20 |
+
"CascadeForCausalLM"
|
| 21 |
+
]
|
| 22 |
+
}
|
checkpoints/step-00010000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fe85b0667957ed5398e54c34aa7d24babc707cff5d635f5af94ba21f8916e7ce
|
| 3 |
+
size 1997546578
|
checkpoints/step-00010000/training_state.json
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"step": 10000,
|
| 3 |
+
"tokens": 983040000,
|
| 4 |
+
"parameters": {
|
| 5 |
+
"total": 998694784,
|
| 6 |
+
"active": 340154240,
|
| 7 |
+
"shared": 10883968,
|
| 8 |
+
"active_path": 329270272
|
| 9 |
+
},
|
| 10 |
+
"batch_size": 96,
|
| 11 |
+
"sequence_length": 1024,
|
| 12 |
+
"loss": 0.99609375,
|
| 13 |
+
"auxiliary_loss": 1.3828125,
|
| 14 |
+
"gradient_norm": 0.1494140625,
|
| 15 |
+
"route": 0,
|
| 16 |
+
"lr_scale": 0.9817758610406299,
|
| 17 |
+
"tokens_per_second": 455814.1952219926,
|
| 18 |
+
"elapsed_seconds": 1598.9329262580013,
|
| 19 |
+
"average_patch_bytes": 8.525932350390287,
|
| 20 |
+
"patch_coalesced_rate": 0.11458333333333333,
|
| 21 |
+
"gate": 0.53515625,
|
| 22 |
+
"peak_vram_gib": 24.91581106185913,
|
| 23 |
+
"documents": {
|
| 24 |
+
"HuggingFaceFW/fineweb-edu:sample-10BT": 73231,
|
| 25 |
+
"HuggingFaceTB/finemath:finemath-4plus": 26090,
|
| 26 |
+
"HuggingFaceTB/smollm-corpus:cosmopedia-v2": 15790,
|
| 27 |
+
"open-r1/OpenR1-Math-220k:default": 32147,
|
| 28 |
+
"open-r1/codeforces-cots:solutions_w_editorials_py_decontaminated": 464
|
| 29 |
+
},
|
| 30 |
+
"source_bytes": {
|
| 31 |
+
"HuggingFaceFW/fineweb-edu:sample-10BT": 342662456,
|
| 32 |
+
"HuggingFaceTB/finemath:finemath-4plus": 129972402,
|
| 33 |
+
"HuggingFaceTB/smollm-corpus:cosmopedia-v2": 59077727,
|
| 34 |
+
"open-r1/OpenR1-Math-220k:default": 41355928,
|
| 35 |
+
"open-r1/codeforces-cots:solutions_w_editorials_py_decontaminated": 17732336
|
| 36 |
+
},
|
| 37 |
+
"validation_loss": 1.064453125
|
| 38 |
+
}
|