Commit
•
a9605db
1
Parent(s):
51273b6
Upload folder using huggingface_hub
Browse files
README.md
CHANGED
|
@@ -217,4 +217,23 @@ The following `bitsandbytes` quantization config was used during training:
|
|
| 217 |
### Framework versions
|
| 218 |
|
| 219 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
- PEFT 0.6.0
|
|
|
|
| 217 |
### Framework versions
|
| 218 |
|
| 219 |
|
| 220 |
+
- PEFT 0.6.0
|
| 221 |
+
## Training procedure
|
| 222 |
+
|
| 223 |
+
|
| 224 |
+
The following `bitsandbytes` quantization config was used during training:
|
| 225 |
+
- quant_method: bitsandbytes
|
| 226 |
+
- load_in_8bit: False
|
| 227 |
+
- load_in_4bit: True
|
| 228 |
+
- llm_int8_threshold: 6.0
|
| 229 |
+
- llm_int8_skip_modules: None
|
| 230 |
+
- llm_int8_enable_fp32_cpu_offload: False
|
| 231 |
+
- llm_int8_has_fp16_weight: False
|
| 232 |
+
- bnb_4bit_quant_type: nf4
|
| 233 |
+
- bnb_4bit_use_double_quant: True
|
| 234 |
+
- bnb_4bit_compute_dtype: float16
|
| 235 |
+
|
| 236 |
+
### Framework versions
|
| 237 |
+
|
| 238 |
+
|
| 239 |
- PEFT 0.6.0
|