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

change bnb tests (#34713)

Browse Source 
* fix training tests

* fix xpu check

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* rm pdb

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* fix 4bit logits check

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* fix 4bit logits check

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* add xpu check on int8 training

* fix training tests

* add llama test on bnb

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* only cpu and xpu disable autocast training

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

* fix format

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

---------

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
Co-authored-by: Titus <9048635+Titus-von-Koeller@users.noreply.github.com>
This commit is contained in:
jiqing-feng
2024-12-18 22:49:59 +08:00
committed by GitHub
parent da334bcfa8
commit 69e31eb1bf
2 changed files with 67 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/quantization/bnb/test_4bit.py
View File

@@ -53,6 +53,8 @@ def get_some_linear_layer(model):
except AttributeError:
# for AutoModelforCausalLM
return model.model.decoder.layers[0].fc1
elif model.config.model_type == "llama":
return model.model.layers[0].mlp.gate_proj
else:
return model.transformer.h[0].mlp.dense_4h_to_h
@@ -106,6 +108,7 @@ class Base4bitTest(unittest.TestCase):
EXPECTED_OUTPUTS.add("Hello my name is John and I am a professional photographer. I")
EXPECTED_OUTPUTS.add("Hello my name is John.\nI am a friend of your father.\n")
EXPECTED_OUTPUTS.add("Hello my name is John Doe, I am a student at the University")
EXPECTED_OUTPUTS.add("Hello my name is John and I am 25 years old.")
MAX_NEW_TOKENS = 10
def setUp(self):
@@ -555,6 +558,8 @@ class Bnb4BitTestTraining(Base4bitTest):
if torch.cuda.is_available():
self.assertEqual(set(model.hf_device_map.values()), {torch.cuda.current_device()})
elif torch.xpu.is_available():
self.assertEqual(set(model.hf_device_map.values()), {f"xpu:{torch.xpu.current_device()}"})
else:
self.assertTrue(all(param.device.type == "cpu" for param in model.parameters()))
@@ -588,11 +593,18 @@ class Bnb4BitTestTraining(Base4bitTest):
@apply_skip_if_not_implemented
@unittest.skipIf(torch_device == "xpu", reason="XPU has precision issue on gpt model, will test it once fixed")
class Bnb4BitGPT2Test(Bnb4BitTest):
model_name = "openai-community/gpt2-xl"
EXPECTED_RELATIVE_DIFFERENCE = 3.3191854854152187
@apply_skip_if_not_implemented
class Bnb4BitLlamaTest(Bnb4BitTest):
model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
EXPECTED_RELATIVE_DIFFERENCE = 2.9461410686392764
@require_bitsandbytes
@require_accelerate
@require_torch
@@ -672,7 +684,7 @@ class BaseSerializationTest(unittest.TestCase):
encoded_input = tokenizer(self.input_text, return_tensors="pt").to(torch_device)
out_0 = model_0(**encoded_input)
out_1 = model_1(**encoded_input)
self.assertTrue(torch.equal(out_0["logits"], out_1["logits"]))
self.assertTrue(torch.allclose(out_0["logits"], out_1["logits"], atol=0.05))
# comparing generate() outputs
encoded_input = tokenizer(self.input_text, return_tensors="pt").to(torch_device)
@@ -734,6 +746,14 @@ class GPTSerializationTest(BaseSerializationTest):
model_name = "openai-community/gpt2-xl"
class LlamaSerializationTest(BaseSerializationTest):
"""
default BaseSerializationTest config tested with Llama family model
"""
model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
@require_bitsandbytes
@require_accelerate
@require_torch_gpu_if_bnb_not_multi_backend_enabled