Commit
•
5d5ee21
1
Parent(s):
a9605db
Upload folder using huggingface_hub
Browse files
README.md
CHANGED
|
@@ -236,4 +236,23 @@ The following `bitsandbytes` quantization config was used during training:
|
|
| 236 |
### Framework versions
|
| 237 |
|
| 238 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 239 |
- PEFT 0.6.0
|
|
|
|
| 236 |
### Framework versions
|
| 237 |
|
| 238 |
|
| 239 |
+
- PEFT 0.6.0
|
| 240 |
+
## Training procedure
|
| 241 |
+
|
| 242 |
+
|
| 243 |
+
The following `bitsandbytes` quantization config was used during training:
|
| 244 |
+
- quant_method: bitsandbytes
|
| 245 |
+
- load_in_8bit: False
|
| 246 |
+
- load_in_4bit: True
|
| 247 |
+
- llm_int8_threshold: 6.0
|
| 248 |
+
- llm_int8_skip_modules: None
|
| 249 |
+
- llm_int8_enable_fp32_cpu_offload: False
|
| 250 |
+
- llm_int8_has_fp16_weight: False
|
| 251 |
+
- bnb_4bit_quant_type: nf4
|
| 252 |
+
- bnb_4bit_use_double_quant: True
|
| 253 |
+
- bnb_4bit_compute_dtype: float16
|
| 254 |
+
|
| 255 |
+
### Framework versions
|
| 256 |
+
|
| 257 |
+
|
| 258 |
- PEFT 0.6.0
|