12777439528

Committed 14 Jan 2025 10:21PM UTC coverage: 95.153% (+14.6%) from 80.557%

Build # 12777439528

Build Type

Pull #62

github

Specific Base 3071c5

Committed by

web-flow

Commit Message

Merge 20548d47d into 42c868843

Pull Request Pull Request #62: feat: Hugging face - speech to text async fix

Run Details

14 of 14 new or added lines in 1 file covered. (100.0%)

33 existing lines in 9 files now uncovered.

2022 of 2125 relevant lines covered (95.15%)

3.8 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

94.55

/scope3ai/tracers/huggingface/translation.py

import time
from dataclasses import dataclass, asdict
from typing import Any, Callable, Optional, Union

import tiktoken
from aiohttp import ClientResponse
from huggingface_hub import InferenceClient, AsyncInferenceClient  # type: ignore[import-untyped]
from huggingface_hub import TranslationOutput as _TranslationOutput
from requests import Response

from scope3ai.api.types import Scope3AIContext, Model, ImpactRow
from scope3ai.api.typesgen import Task
from scope3ai.constants import PROVIDERS
from scope3ai.lib import Scope3AI
from scope3ai.response_interceptor.aiohttp_interceptor import aiohttp_response_capture
from scope3ai.response_interceptor.requests_interceptor import requests_response_capture

PROVIDER = PROVIDERS.HUGGINGFACE_HUB.value
HUGGING_FACE_TRANSLATION_TASK = "translation"


@dataclass
class TranslationOutput(_TranslationOutput):
    scope3ai: Optional[Scope3AIContext] = None


def _hugging_face_translation_wrapper(
    timer_start: Any,
    model: Any,
    response: Any,
    http_response: Union[ClientResponse, Response],
    args: Any,
    kwargs: Any,
) -> TranslationOutput:
    encoder = tiktoken.get_encoding("cl100k_base")
    input_tokens = None
    if http_response:
        compute_time = http_response.headers.get("x-compute-time")
        input_tokens = http_response.headers.get("x-compute-characters")
    else:
        compute_time = time.perf_counter() - timer_start
    if not input_tokens:
        prompt = args[0] if len(args) > 0 else kwargs.get("text", "")
        input_tokens = len(encoder.encode(prompt)) if prompt != "" else 0
    output_tokens = len(encoder.encode(response.translation_text))
    scope3_row = ImpactRow(
        model=Model(id=model),
        task=Task.translation,
        input_tokens=input_tokens,
        output_tokens=output_tokens,
        request_duration_ms=float(compute_time) * 1000,
        managed_service_id=PROVIDER,
    )

    scope3_ctx = Scope3AI.get_instance().submit_impact(scope3_row)
    result = TranslationOutput(**asdict(response))
    result.scope3ai = scope3_ctx
    return result


async def huggingface_translation_wrapper_async_non_stream(
    wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
) -> TranslationOutput:
    timer_start = time.perf_counter()
    http_response: ClientResponse | None = None
    with aiohttp_response_capture() as responses:
        response = await wrapped(*args, **kwargs)
        http_responses = responses.get()
        if len(http_responses) > 0:
            http_response = http_responses[-1]
    model = kwargs.get("model") or instance.get_recommended_model(
        HUGGING_FACE_TRANSLATION_TASK
    )
    return _hugging_face_translation_wrapper(
        timer_start, model, response, http_response, args, kwargs
    )


def huggingface_translation_wrapper_non_stream(
    wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
) -> TranslationOutput:
    timer_start = time.perf_counter()
    http_response: Response | None = None
    with requests_response_capture() as responses:
        response = wrapped(*args, **kwargs)
        http_responses = responses.get()
        if len(http_responses) > 0:
            http_response = http_responses[-1]
    model = kwargs.get("model") or instance.get_recommended_model(
        HUGGING_FACE_TRANSLATION_TASK
    )
    return _hugging_face_translation_wrapper(
        timer_start, model, response, http_response, args, kwargs
    )

1	import time	4✔
2	from dataclasses import dataclass, asdict	4✔
3	from typing import Any, Callable, Optional, Union	4✔
4
5	import tiktoken	4✔
6	from aiohttp import ClientResponse	4✔
7	from huggingface_hub import InferenceClient, AsyncInferenceClient # type: ignore[import-untyped]	4✔
8	from huggingface_hub import TranslationOutput as _TranslationOutput	4✔
9	from requests import Response	4✔
10
11	from scope3ai.api.types import Scope3AIContext, Model, ImpactRow	4✔
12	from scope3ai.api.typesgen import Task	4✔
13	from scope3ai.constants import PROVIDERS	4✔
14	from scope3ai.lib import Scope3AI	4✔
15	from scope3ai.response_interceptor.aiohttp_interceptor import aiohttp_response_capture	4✔
16	from scope3ai.response_interceptor.requests_interceptor import requests_response_capture	4✔
17
18	PROVIDER = PROVIDERS.HUGGINGFACE_HUB.value	4✔
19	HUGGING_FACE_TRANSLATION_TASK = "translation"	4✔
20
21
22	@dataclass	4✔
23	class TranslationOutput(_TranslationOutput):	4✔
24	scope3ai: Optional[Scope3AIContext] = None	4✔
25
26
27	def _hugging_face_translation_wrapper(	4✔
28	timer_start: Any,
29	model: Any,
30	response: Any,
31	http_response: Union[ClientResponse, Response],
32	args: Any,
33	kwargs: Any,
34	) -> TranslationOutput:
35	encoder = tiktoken.get_encoding("cl100k_base")	4✔
36	input_tokens = None	4✔
37	if http_response:	4✔
38	compute_time = http_response.headers.get("x-compute-time")	4✔
39	input_tokens = http_response.headers.get("x-compute-characters")	4✔
40	else:
UNCOV 41	compute_time = time.perf_counter() - timer_start	×
42	if not input_tokens:	4✔
UNCOV 43	prompt = args[0] if len(args) > 0 else kwargs.get("text", "")	×
UNCOV 44	input_tokens = len(encoder.encode(prompt)) if prompt != "" else 0	×
45	output_tokens = len(encoder.encode(response.translation_text))	4✔
46	scope3_row = ImpactRow(	4✔
47	model=Model(id=model),
48	task=Task.translation,
49	input_tokens=input_tokens,
50	output_tokens=output_tokens,
51	request_duration_ms=float(compute_time) * 1000,
52	managed_service_id=PROVIDER,
53	)
54
55	scope3_ctx = Scope3AI.get_instance().submit_impact(scope3_row)	4✔
56	result = TranslationOutput(**asdict(response))	4✔
57	result.scope3ai = scope3_ctx	4✔
58	return result	4✔
59
60
61	async def huggingface_translation_wrapper_async_non_stream(	4✔
62	wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
63	) -> TranslationOutput:
64	timer_start = time.perf_counter()	4✔
65	http_response: ClientResponse \| None = None	4✔
66	with aiohttp_response_capture() as responses:	4✔
67	response = await wrapped(args, *kwargs)	4✔
68	http_responses = responses.get()	4✔
69	if len(http_responses) > 0:	4✔
70	http_response = http_responses[-1]	4✔
71	model = kwargs.get("model") or instance.get_recommended_model(	4✔
72	HUGGING_FACE_TRANSLATION_TASK
73	)
74	return _hugging_face_translation_wrapper(	4✔
75	timer_start, model, response, http_response, args, kwargs
76	)
77
78
79	def huggingface_translation_wrapper_non_stream(	4✔
80	wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
81	) -> TranslationOutput:
82	timer_start = time.perf_counter()	4✔
83	http_response: Response \| None = None	4✔
84	with requests_response_capture() as responses:	4✔
85	response = wrapped(args, *kwargs)	4✔
86	http_responses = responses.get()	4✔
87	if len(http_responses) > 0:	4✔
88	http_response = http_responses[-1]	4✔
89	model = kwargs.get("model") or instance.get_recommended_model(	4✔
90	HUGGING_FACE_TRANSLATION_TASK
91	)
92	return _hugging_face_translation_wrapper(	4✔
93	timer_start, model, response, http_response, args, kwargs
94	)

scope3data / scope3ai-py / 12777439528

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous