Instructions to use guygrigsby/diff-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use guygrigsby/diff-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir diff-mlx guygrigsby/diff-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
Upload vanilla/config.json with huggingface_hub
Browse files- vanilla/config.json +32 -0
vanilla/config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"dim": 768,
|
| 4 |
+
"n_layers": 12,
|
| 5 |
+
"n_heads_vanilla": 12,
|
| 6 |
+
"qk_head_dim": 64,
|
| 7 |
+
"vocab_size": 100277,
|
| 8 |
+
"mlp_intermediate": 2048,
|
| 9 |
+
"block_size": 2048,
|
| 10 |
+
"rope_base": 10000.0,
|
| 11 |
+
"rms_eps": 1e-05,
|
| 12 |
+
"tie_embeddings": true,
|
| 13 |
+
"amp_dtype": "bfloat16"
|
| 14 |
+
},
|
| 15 |
+
"train": {
|
| 16 |
+
"peak_lr": 0.0004,
|
| 17 |
+
"warmup_steps": 1000,
|
| 18 |
+
"total_tokens": 2000000000,
|
| 19 |
+
"micro_batch": 8,
|
| 20 |
+
"grad_accum": 4,
|
| 21 |
+
"weight_decay": 0.1,
|
| 22 |
+
"adam_beta1": 0.9,
|
| 23 |
+
"adam_beta2": 0.95,
|
| 24 |
+
"adam_eps": 1e-08,
|
| 25 |
+
"grad_clip": 1.0,
|
| 26 |
+
"eval_every": 1000,
|
| 27 |
+
"full_eval_every": 5000,
|
| 28 |
+
"monitoring_tokens": 2000000,
|
| 29 |
+
"full_eval_tokens": 75000000,
|
| 30 |
+
"save_every": 500
|
| 31 |
+
}
|
| 32 |
+
}
|