huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-04-27 13:02:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: update mamba snap and run other release tests

Browse Source
This commit is contained in:
drbh 2024-08-08 17:55:24 +00:00
parent add7908f0f
commit 700e64c5b9
3 changed files with 18 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
integration-tests/models/__snapshots__/test_mamba/test_mamba_all_params.json
View File

@ -11,22 +11,22 @@
}, },
{ {
"id": 13, "id": 13,
"logprob": -2.734375, "logprob": -2.59375,
"text": "," "text": ","
}, },
{ {
"id": 8862, "id": 8862,
"logprob": -3.6875, "logprob": -3.5625,
"text": " yellow" "text": " yellow"
}, },
{ {
"id": 13, "id": 13,
"logprob": -0.40234375, "logprob": -0.44726562,
"text": "," "text": ","
}, },
{ {
"id": 209, "id": 209,
"logprob": -8.25, "logprob": -8.0,
"text": " " "text": " "
} }
], ],
@ -52,7 +52,7 @@
}, },
{ {
"id": 9830, "id": 9830,
"logprob": -2.25, "logprob": -2.03125,
"special": false, "special": false,
"text": " colors" "text": " colors"
}, },
@ -64,13 +64,13 @@
}, },
{ {
"id": 329, "id": 329,
"logprob": -2.171875, "logprob": -2.734375,
"special": false, "special": false,
"text": " A" "text": " A"
}, },
{ {
"id": 1180, "id": 1180,
"logprob": -2.046875, "logprob": -2.0,
"special": false, "special": false,
"text": " number" "text": " number"
}, },
@ -81,19 +81,19 @@
"text": " of" "text": " of"
}, },
{ {
"id": 1027, "id": 253,
"logprob": -1.5546875, "logprob": -0.69140625,
"special": false, "special": false,
"text": " different" "text": " the"
}, },
{ {
"id": 3295, "id": 3295,
"logprob": -0.97265625, "logprob": -0.8203125,
"special": false, "special": false,
"text": " color" "text": " color"
} }
], ],
"top_tokens": null "top_tokens": null
}, },
"generated_text": "blue, red, yellow, \nand blue colors. A number of different color" "generated_text": "blue, red, yellow, \nand blue colors. A number of the color"
} }

6
integration-tests/models/test_flash_llama_exl2.py
View File

@ -21,7 +21,7 @@ async def flash_llama_exl2(flash_llama_exl2_handle):
return flash_llama_exl2_handle.client return flash_llama_exl2_handle.client
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_exl2(flash_llama_exl2, ignore_logprob_response_snapshot): async def test_flash_llama_exl2(flash_llama_exl2, ignore_logprob_response_snapshot):
@ -33,7 +33,7 @@ async def test_flash_llama_exl2(flash_llama_exl2, ignore_logprob_response_snapsh
assert response == ignore_logprob_response_snapshot assert response == ignore_logprob_response_snapshot
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_exl2_all_params( async def test_flash_llama_exl2_all_params(
@ -60,7 +60,7 @@ async def test_flash_llama_exl2_all_params(
assert response == ignore_logprob_response_snapshot assert response == ignore_logprob_response_snapshot
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_exl2_load( async def test_flash_llama_exl2_load(

6
integration-tests/models/test_flash_llama_fp8.py
View File

@ -13,7 +13,7 @@ async def flash_llama_fp8(flash_llama_fp8_handle):
return flash_llama_fp8_handle.client return flash_llama_fp8_handle.client
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_fp8(flash_llama_fp8, response_snapshot): async def test_flash_llama_fp8(flash_llama_fp8, response_snapshot):
@ -25,7 +25,7 @@ async def test_flash_llama_fp8(flash_llama_fp8, response_snapshot):
assert response == response_snapshot assert response == response_snapshot
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_fp8_all_params(flash_llama_fp8, response_snapshot): async def test_flash_llama_fp8_all_params(flash_llama_fp8, response_snapshot):
@ -48,7 +48,7 @@ async def test_flash_llama_fp8_all_params(flash_llama_fp8, response_snapshot):
assert response == response_snapshot assert response == response_snapshot
@pytest.mark.release # @pytest.mark.release
@pytest.mark.asyncio @pytest.mark.asyncio
@pytest.mark.private @pytest.mark.private
async def test_flash_llama_fp8_load(flash_llama_fp8, generate_load, response_snapshot): async def test_flash_llama_fp8_load(flash_llama_fp8, generate_load, response_snapshot):