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

fix(test): replace jackfram llama with smollm

Browse Source
This commit is contained in:
NanoCode012
2025-02-28 16:40:49 +07:00
parent 75cbd15301
commit 10d18e6c97
12 changed files with 26 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/e2e/patched/test_4d_multipack_llama.py
View File

@@ -28,7 +28,7 @@ class Test4dMultipackLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"flash_attention": False,
"sdp_attention": True,
"sample_packing": True,
@@ -72,7 +72,7 @@ class Test4dMultipackLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"flash_attention": False,
"sdp_attention": False,
"sample_packing": True,

2
tests/e2e/patched/test_fused_llama.py
View File

@@ -32,7 +32,7 @@ class TestFusedLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"flash_attention": True,
"pad_to_sequence_len": True,
"flash_attn_fuse_qkv": True,

6
tests/e2e/patched/test_llama_s2_attention.py
View File

@@ -31,8 +31,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"sequence_len": 16384,
"sample_packing": False,
"flash_attention": True,
@@ -77,8 +76,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"sequence_len": 16384,
"sample_packing": False,
"flash_attention": True,

4
tests/e2e/patched/test_lora_llama_multipack.py
View File

@@ -31,8 +31,7 @@ class TestLoraLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
"sequence_len": 1024,
"sample_packing": True,
"flash_attention": True,
@@ -43,6 +42,7 @@ class TestLoraLlama(unittest.TestCase):
"lora_dropout": 0.05,
"lora_target_linear": True,
"val_set_size": 0.2,
"lora_modules_to_save": ["lm_head", "embed_tokens"],
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",