Instructions to use nahommohan/tibeb-sft-adapter with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use nahommohan/tibeb-sft-adapter with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir tibeb-sft-adapter nahommohan/tibeb-sft-adapter
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Upload folder using huggingface_hub
Browse files
mlx-adapter-v3/0001000_adapters.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:da06abaf4901705d141430c3c7df98cb43c9c501b424f8bebbbf00f35e3a84aa
|
| 3 |
+
size 167821465
|
mlx-adapter-v3/0002000_adapters.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:78fb08cd3dc6537186c2528d58fe1cdfd34ec2c768df8ea62d87220a9b4a095e
|
| 3 |
+
size 167821465
|
mlx-adapter-v3/adapter_config.json
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"adapter_path": "models/tibeb-sft/mlx-adapter-v3",
|
| 3 |
+
"batch_size": 1,
|
| 4 |
+
"config": "configs/v3_config.yaml",
|
| 5 |
+
"data": "data/mlx_train",
|
| 6 |
+
"fine_tune_type": "lora",
|
| 7 |
+
"grad_accumulation_steps": 4,
|
| 8 |
+
"grad_checkpoint": true,
|
| 9 |
+
"iters": 20000,
|
| 10 |
+
"learning_rate": 1e-05,
|
| 11 |
+
"lora_parameters": {
|
| 12 |
+
"rank": 16,
|
| 13 |
+
"dropout": 0.05,
|
| 14 |
+
"scale": 32.0
|
| 15 |
+
},
|
| 16 |
+
"lr_schedule": {
|
| 17 |
+
"name": "cosine_decay",
|
| 18 |
+
"arguments": [
|
| 19 |
+
1e-05,
|
| 20 |
+
20000,
|
| 21 |
+
1e-06
|
| 22 |
+
],
|
| 23 |
+
"warmup": 500,
|
| 24 |
+
"warmup_init": 1e-07
|
| 25 |
+
},
|
| 26 |
+
"mask_prompt": false,
|
| 27 |
+
"max_seq_length": 512,
|
| 28 |
+
"model": "mlx-community/aya-expanse-8b-4bit",
|
| 29 |
+
"num_layers": -1,
|
| 30 |
+
"optimizer": "adamw",
|
| 31 |
+
"optimizer_config": {
|
| 32 |
+
"adam": {},
|
| 33 |
+
"adamw": {},
|
| 34 |
+
"muon": {},
|
| 35 |
+
"sgd": {},
|
| 36 |
+
"adafactor": {}
|
| 37 |
+
},
|
| 38 |
+
"project_name": null,
|
| 39 |
+
"report_to": null,
|
| 40 |
+
"resume_adapter_file": null,
|
| 41 |
+
"save_every": 1000,
|
| 42 |
+
"seed": 42,
|
| 43 |
+
"steps_per_eval": 500,
|
| 44 |
+
"steps_per_report": 10,
|
| 45 |
+
"test": false,
|
| 46 |
+
"test_batches": 500,
|
| 47 |
+
"train": true,
|
| 48 |
+
"val_batches": 25
|
| 49 |
+
}
|
mlx-adapter-v3/adapters.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:78fb08cd3dc6537186c2528d58fe1cdfd34ec2c768df8ea62d87220a9b4a095e
|
| 3 |
+
size 167821465
|