SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix quantization tests (#29914)

Browse Source 
* revert back to torch 2.1.1

* run test

* switch to torch 2.2.1

* udapte dockerfile

* fix awq tests

* fix test

* run quanto tests

* update tests

* split quantization tests

* fix

* fix again

* final fix

* fix report artifact

* build docker again

* Revert "build docker again"

This reverts commit 399a5f9d9308da071d79034f238c719de0f3532e.

* debug

* revert

* style

* new notification system

* testing notfication

* rebuild docker

* fix_prev_ci_results

* typo

* remove warning

* fix typo

* fix artifact name

* debug

* issue fixed

* debug again

* fix

* fix time

* test notif with faling test

* typo

* issues again

* final fix ?

* run all quantization tests again

* remove name to clear space

* revert modfiication done on workflow

* fix

* build docker

* build only quant docker

* fix quantization ci

* fix

* fix report

* better quantization_matrix

* add print

* revert to the basic one
This commit is contained in:
Marc Sun
2024-04-09 17:10:29 +02:00
committed by GitHub
parent 6487e9b370
commit 58a939c6b7
7 changed files with 324 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
tests/quantization/autoawq/test_awq.py
View File

@@ -101,7 +101,7 @@ class AwqTest(unittest.TestCase):
EXPECTED_OUTPUT = "Hello my name is Katie and I am a 20 year old student at the University of North Carolina at Chapel Hill. I am a junior and I am majoring in Journalism and minoring in Spanish"
EXPECTED_OUTPUT_BF16 = "Hello my name is Katie and I am a 20 year old student at the University of North Carolina at Chapel Hill. I am a junior and I am majoring in Exercise and Sport Science with a"
EXPECTED_OUTPUT_EXLLAMA = "Hello my name is Katie and I am a 20 year old student from the UK. I am currently studying for a degree in English Literature and History at the University of York. I am a very out"
device_map = "cuda"
# called only once for all test in this class
@@ -200,11 +200,11 @@ class AwqTest(unittest.TestCase):
quantization_config = AwqConfig(version="exllama")
quantized_model = AutoModelForCausalLM.from_pretrained(
self.model_name, quantization_config=quantization_config
).to(torch_device)
self.model_name, quantization_config=quantization_config, device_map=torch_device
)
output = quantized_model.generate(**input_ids, max_new_tokens=40)
self.assertEqual(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUT)
self.assertEqual(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUT_EXLLAMA)
def test_quantized_model_no_device_map(self):
"""
@@ -239,7 +239,7 @@ class AwqTest(unittest.TestCase):
quantized_model = AutoModelForCausalLM.from_pretrained(self.model_name, device_map="auto")
self.assertTrue(set(quantized_model.hf_device_map.values()) == {0, 1, 2, 3})
self.assertTrue(set(quantized_model.hf_device_map.values()) == {0, 1})
output = quantized_model.generate(**input_ids, max_new_tokens=40)
@@ -272,8 +272,8 @@ class AwqFusedTest(unittest.TestCase):
model_name = "TheBloke/Mistral-7B-OpenOrca-AWQ"
model_revision = "7048b2af77d0dd1c81b000b19d73f9cc8950b510"
custom_mapping_model_id = "TheBloke/Yi-34B-AWQ"
custom_model_revision = "f1b2cd1b7459ceecfdc1fac5bb8725f13707c589"
custom_mapping_model_id = "TheBloke/Mistral-7B-v0.1-AWQ"
custom_model_revision = "f186bcfa9edbe2a4334262ec1e67f23e53ed1ae7"
mixtral_model_name = "casperhansen/mixtral-instruct-awq"
mixtral_model_revision = "87dd4ec502dde74fb3a624835c776b000d190c3b"
@@ -287,8 +287,8 @@ class AwqFusedTest(unittest.TestCase):
"You end up exactly where you started. Where are you?"
)
EXPECTED_GENERATION = prompt + "\n\nThis is a classic puzzle that has been around for"
EXPECTED_GENERATION_CUSTOM_MODEL = "HelloWorld.java:11)\r\n\tat org"
EXPECTED_GENERATION = prompt + "\n\nYou are at the starting point.\n\nIf"
EXPECTED_GENERATION_CUSTOM_MODEL = "Hello,\n\nI have a problem with my 20"
EXPECTED_GENERATION_MIXTRAL = prompt + " You're on the North Pole.\n\nThe"
def tearDown(self):
@@ -423,28 +423,25 @@ class AwqFusedTest(unittest.TestCase):
fuse_max_seq_len=512,
modules_to_fuse={
"attention": ["q_proj", "k_proj", "v_proj", "o_proj"],
"layernorm": ["ln1", "ln2", "norm"],
"mlp": ["gate_proj", "up_proj", "down_proj"],
"layernorm": ["input_layernorm", "post_attention_layernorm", "norm"],
"use_alibi": False,
"num_attention_heads": 56,
"hidden_size": 4096,
"num_attention_heads": 32,
"num_key_value_heads": 8,
"hidden_size": 7168,
},
)
model = AutoModelForCausalLM.from_pretrained(
self.custom_mapping_model_id,
quantization_config=quantization_config,
trust_remote_code=True,
device_map="balanced",
revision=self.custom_model_revision,
)
self._check_fused_modules(model)
tokenizer = AutoTokenizer.from_pretrained(
self.custom_mapping_model_id, revision=self.custom_model_revision, trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(self.custom_mapping_model_id, revision=self.custom_model_revision)
prompt = "Hello"
inputs = tokenizer(prompt, return_tensors="pt").to(torch_device)
@@ -452,6 +449,7 @@ class AwqFusedTest(unittest.TestCase):
outputs = model.generate(**inputs, max_new_tokens=12)
self.assertEqual(tokenizer.decode(outputs[0], skip_special_tokens=True), self.EXPECTED_GENERATION_CUSTOM_MODEL)
@unittest.skip("Not enough GPU memory on CI runners")
@require_torch_multi_gpu
def test_generation_mixtral_fused(self):
"""