Spaces:
Sleeping
Sleeping
Commit ·
8da3892
1
Parent(s): 13bffd9
Upload folder using huggingface_hub
Browse files
training/MaterialForge_GRPO_Training.ipynb
CHANGED
|
@@ -357,7 +357,7 @@
|
|
| 357 |
"source": [
|
| 358 |
"## 7. Load Model with Unsloth\n",
|
| 359 |
"\n",
|
| 360 |
-
"We use Unsloth for memory-efficient 4-bit QLoRA fine-tuning. This notebook now defaults to `Qwen/Qwen3-
|
| 361 |
]
|
| 362 |
},
|
| 363 |
{
|
|
@@ -508,7 +508,7 @@
|
|
| 508 |
"import torch\n",
|
| 509 |
"from unsloth import FastLanguageModel\n",
|
| 510 |
"\n",
|
| 511 |
-
"MODEL_NAME = os.getenv(\"MATERIALFORGE_MODEL_NAME\", \"Qwen/Qwen3-
|
| 512 |
"MAX_SEQ_LENGTH = 4096\n",
|
| 513 |
"\n",
|
| 514 |
"model, tokenizer = FastLanguageModel.from_pretrained(\n",
|
|
|
|
| 357 |
"source": [
|
| 358 |
"## 7. Load Model with Unsloth\n",
|
| 359 |
"\n",
|
| 360 |
+
"We use Unsloth for memory-efficient 4-bit QLoRA fine-tuning. This notebook now defaults to `Qwen/Qwen3-14B` for a stronger tool-using policy on larger hardware. If you want to try another checkpoint, override the model name with an environment variable instead of editing the notebook repeatedly."
|
| 361 |
]
|
| 362 |
},
|
| 363 |
{
|
|
|
|
| 508 |
"import torch\n",
|
| 509 |
"from unsloth import FastLanguageModel\n",
|
| 510 |
"\n",
|
| 511 |
+
"MODEL_NAME = os.getenv(\"MATERIALFORGE_MODEL_NAME\", \"Qwen/Qwen3-14B\")\n",
|
| 512 |
"MAX_SEQ_LENGTH = 4096\n",
|
| 513 |
"\n",
|
| 514 |
"model, tokenizer = FastLanguageModel.from_pretrained(\n",
|