mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-20 06:12:07 +00:00
ba291dad9f
* Improve the handling of quantized weights Handling of quantized weights was split between two mechanisms: - For quantized checkpoints, we used the new weight loader infrastructure. - For quantization while loading (EETQ, FP8, bitsandbytes) we instead relied on conditional in `get_linear`. Weight loaders support context managers to selectively load particular layers with different weight loaders, which is useful for models like Idefics2 AWQ, which uses a quantized text model, but unquantized vision and connector models. However, the context manager would be overrided by `get_linear`, which string-checks `quantizer`. Also, the context manager would not work with EETQ, FP8, and bitsandbytes. This change migrates all quantizers to the weight loader infrastructure. This has several benefits: - We can use context managers with all quantizers. - All the implementation details move down to the quantizer layers, `get_linear` does not need to know how to handle quantizer linear layers. - All quantizer weights are strongly typed, we don't pass around raw tensors. - We don't have to pass around the `quantizer` string everywhere. * Exclude non-MLP layers when using FP8 quantization with Llama
67 lines
1.8 KiB
Python
67 lines
1.8 KiB
Python
import pytest
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def flash_llama_marlin24_handle(launcher):
|
|
with launcher(
|
|
"nm-testing/Llama-2-7b-pruned2.4-Marlin_24", quantize="marlin"
|
|
) as handle:
|
|
yield handle
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
async def flash_llama_marlin(flash_llama_marlin24_handle):
|
|
await flash_llama_marlin24_handle.health(300)
|
|
return flash_llama_marlin24_handle.client
|
|
|
|
|
|
@pytest.mark.release
|
|
@pytest.mark.asyncio
|
|
@pytest.mark.private
|
|
async def test_flash_llama_marlin(flash_llama_marlin, response_snapshot):
|
|
response = await flash_llama_marlin.generate(
|
|
"Test request", max_new_tokens=10, decoder_input_details=True
|
|
)
|
|
|
|
assert response.details.generated_tokens == 10
|
|
assert response == response_snapshot
|
|
|
|
|
|
@pytest.mark.release
|
|
@pytest.mark.asyncio
|
|
@pytest.mark.private
|
|
async def test_flash_llama_marlin24_all_params(flash_llama_marlin, response_snapshot):
|
|
response = await flash_llama_marlin.generate(
|
|
"Test request",
|
|
max_new_tokens=10,
|
|
repetition_penalty=1.2,
|
|
return_full_text=True,
|
|
temperature=0.5,
|
|
top_p=0.9,
|
|
top_k=10,
|
|
truncate=5,
|
|
typical_p=0.9,
|
|
watermark=True,
|
|
decoder_input_details=True,
|
|
seed=0,
|
|
)
|
|
|
|
assert response.details.generated_tokens == 10
|
|
assert response == response_snapshot
|
|
|
|
|
|
@pytest.mark.release
|
|
@pytest.mark.asyncio
|
|
@pytest.mark.private
|
|
async def test_flash_llama_marlin24_load(
|
|
flash_llama_marlin, generate_load, response_snapshot
|
|
):
|
|
responses = await generate_load(
|
|
flash_llama_marlin, "Test request", max_new_tokens=10, n=4
|
|
)
|
|
|
|
assert len(responses) == 4
|
|
assert all([r.generated_text == responses[0].generated_text for r in responses])
|
|
|
|
assert responses == response_snapshot
|