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

replace references to random 68m model w 135m smollm2 (#2570) [skip ci]

Browse Source 
* replace references to random 68m model w 135m smollm2

* use AutoTokenizer for smollm2
This commit is contained in:
Wing Lian
2025-04-28 10:08:07 -04:00
committed by GitHub
parent f1df73a798
commit 40f4ea23ab
13 changed files with 95 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
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",
"flash_attention": False,
"sdp_attention": True,
"sample_packing": True,
@@ -41,6 +41,9 @@ class Test4dMultipackLlama(unittest.TestCase):
"lora_target_linear": True,
"sequence_len": 1024,
"val_set_size": 0.02,
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
@@ -73,7 +76,7 @@ class Test4dMultipackLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"flash_attention": False,
"sdp_attention": False,
"sample_packing": True,
@@ -86,6 +89,9 @@ class Test4dMultipackLlama(unittest.TestCase):
"lora_dropout": 0.05,
"lora_target_linear": True,
"val_set_size": 0.02,
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",

6
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",
"flash_attention": True,
"pad_to_sequence_len": True,
"flash_attn_fuse_qkv": True,
@@ -41,9 +41,7 @@ class TestFusedLlama(unittest.TestCase):
"sequence_len": 1024,
"val_set_size": 0.02,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<|endoftext|>",
},
"datasets": [
{

16
tests/e2e/patched/test_llama_s2_attention.py
View File

@@ -31,8 +31,8 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 16384,
"sample_packing": False,
"flash_attention": True,
@@ -44,7 +44,9 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
"lora_dropout": 0.05,
"lora_target_linear": True,
"val_set_size": 0.02,
"special_tokens": {},
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "Yukang/LongAlpaca-12k",
@@ -78,14 +80,16 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 16384,
"sample_packing": False,
"flash_attention": True,
"s2_attention": True,
"val_set_size": 0.02,
"special_tokens": {},
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "Yukang/LongAlpaca-12k",

16
tests/e2e/patched/test_lora_llama_multipack.py
View File

@@ -31,8 +31,8 @@ class TestLoraLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 1024,
"sample_packing": True,
"flash_attention": True,
@@ -44,9 +44,7 @@ class TestLoraLlama(unittest.TestCase):
"lora_target_linear": True,
"val_set_size": 0.2,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<|endoftext|>",
},
"datasets": [
{
@@ -84,9 +82,9 @@ class TestLoraLlama(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "TheBlokeAI/jackfram_llama-68m-GPTQ",
"base_model": "lilmeaty/SmolLM2-135M-Instruct-GPTQ",
"model_type": "AutoModelForCausalLM",
"tokenizer_type": "LlamaTokenizer",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 1024,
"sample_packing": True,
"flash_attention": True,
@@ -100,9 +98,7 @@ class TestLoraLlama(unittest.TestCase):
"lora_target_linear": True,
"val_set_size": 0.02,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<|endoftext|>",
},
"datasets": [
{