tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

update multigpu/test_eval, multigpu/test_llama

Browse Source
This commit is contained in:
Sunny Liu
2025-04-21 16:49:08 -04:00
parent a20e753321
commit 1aa9f7d952
2 changed files with 32 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/e2e/multigpu/test_eval.py
View File

@@ -30,8 +30,10 @@ class TestMultiGPUEval:
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM2-135M",
"load_in_8bit": False,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
"strict": False,
"sequence_len": 2048,
"adapter": "qlora",
@@ -99,8 +101,10 @@ class TestMultiGPUEval:
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM2-135M",
"load_in_8bit": False,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
"strict": False,
"sequence_len": 2048,
"adapter": "qlora",

30
tests/e2e/multigpu/test_llama.py
View File

@@ -171,7 +171,10 @@ class TestMultiGPULlama:
"sample_packing": False,
"eval_sample_packing": False,
"pad_to_sequence_len": True,
"load_in_8bit": True,
"quantization": {
"backend": "bnb",
"bits": 8,
},
"adapter": "lora",
"lora_r": 8,
"lora_alpha": 16,
@@ -249,7 +252,10 @@ class TestMultiGPULlama:
"sample_packing": False,
"eval_sample_packing": False,
"pad_to_sequence_len": True,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
"adapter": "qlora",
"lora_r": 8,
"lora_alpha": 16,
@@ -548,7 +554,10 @@ class TestMultiGPULlama:
"base_model": "axolotl-ai-co/SmolLM2-135M-bnb-nf4-bf16",
"adapter": "qlora",
"mean_resizing_embeddings": True,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
"lora_r": 8,
"lora_alpha": 16,
"lora_dropout": 0.05,
@@ -648,7 +657,10 @@ class TestMultiGPULlama:
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_linear": True,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
}
else:
adapter = {}
@@ -722,7 +734,10 @@ class TestMultiGPULlama:
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_linear": True,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
}
else:
adapter = {}
@@ -796,7 +811,10 @@ class TestMultiGPULlama:
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_linear": True,
"load_in_4bit": True,
"quantization": {
"backend": "bnb",
"bits": 4,
},
}
else:
adapter = {}