13396554446

import time
from collections.abc import AsyncIterable, Iterable
from dataclasses import asdict, dataclass
from typing import Any, Callable, Optional, Union

import tiktoken
from huggingface_hub import (  # type: ignore[import-untyped]
    AsyncInferenceClient,
    InferenceClient,
)
from huggingface_hub import ChatCompletionOutput as _ChatCompletionOutput
from huggingface_hub import ChatCompletionStreamOutput as _ChatCompletionStreamOutput
from requests import Response

from scope3ai.api.types import ImpactRow, Scope3AIContext
from scope3ai.lib import Scope3AI
from scope3ai.response_interceptor.requests_interceptor import requests_response_capture

HUGGING_FACE_CHAT_TASK = "chat"


@dataclass
class ChatCompletionOutput(_ChatCompletionOutput):
    scope3ai: Optional[Scope3AIContext] = None


@dataclass
class ChatCompletionStreamOutput(_ChatCompletionStreamOutput):
    scope3ai: Optional[Scope3AIContext] = None


def huggingface_chat_wrapper(
    wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
) -> Union[ChatCompletionOutput, Iterable[ChatCompletionStreamOutput]]:
    if kwargs.get("stream", False):
        return huggingface_chat_wrapper_stream(wrapped, instance, args, kwargs)
    else:
        return huggingface_chat_wrapper_non_stream(wrapped, instance, args, kwargs)


def huggingface_chat_wrapper_non_stream(
    wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
) -> ChatCompletionOutput:
    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 http_responses:
            http_response = http_responses[0]
    model = (
        instance.model
        or kwargs.get("model")
        or instance.get_recommended_model(HUGGING_FACE_CHAT_TASK)
    )
    if http_response:
        compute_time = http_response.headers.get("x-compute-time")
    else:
        compute_time = time.perf_counter() - timer_start
    scope3_row = ImpactRow(
        model_id=model,
        input_tokens=response.usage.prompt_tokens,
        output_tokens=response.usage.completion_tokens,
        request_duration_ms=float(compute_time) * 1000,
    )
    scope3ai_ctx = Scope3AI.get_instance().submit_impact(scope3_row)
    chat = ChatCompletionOutput(**asdict(response))
    chat.scope3ai = scope3ai_ctx
    return chat


def huggingface_chat_wrapper_stream(
    wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
) -> Iterable[ChatCompletionStreamOutput]:
    timer_start = time.perf_counter()
    stream = wrapped(*args, **kwargs)
    token_count = 0
    model = (
        instance.model
        or kwargs.get("model")
        or instance.get_recommended_model(HUGGING_FACE_CHAT_TASK)
    )
    for chunk in stream:
        token_count += 1
        request_latency = time.perf_counter() - timer_start
        scope3_row = ImpactRow(
            model_id=model,
            output_tokens=token_count,
            request_duration_ms=request_latency * 1000,
        )
        chunk_data = ChatCompletionStreamOutput(**asdict(chunk))
        scope3_ctx = Scope3AI.get_instance().submit_impact(scope3_row)
        if scope3_ctx is not None:
            chunk_data.scope3ai = scope3_ctx
        yield chunk_data


async def huggingface_async_chat_wrapper(
    wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
) -> Union[ChatCompletionOutput, AsyncIterable[ChatCompletionStreamOutput]]:
    if kwargs.get("stream", False):
        return huggingface_async_chat_wrapper_stream(wrapped, instance, args, kwargs)
    else:
        return await huggingface_async_chat_wrapper_non_stream(
            wrapped, instance, args, kwargs
        )


async def huggingface_async_chat_wrapper_non_stream(
    wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
) -> ChatCompletionOutput:
    timer_start = time.perf_counter()

    response = await wrapped(*args, **kwargs)
    request_latency = time.perf_counter() - timer_start
    model = (
        instance.model or kwargs.get("model") or instance.get_recommended_model("chat")
    )
    encoder = tiktoken.get_encoding("cl100k_base")
    output_tokens = len(encoder.encode(response.choices[0].message.content))
    scope3_row = ImpactRow(
        model_id=model,
        input_tokens=response.usage.prompt_tokens,
        output_tokens=output_tokens,
        request_duration_ms=request_latency * 1000,
    )

    scope3ai_ctx = await Scope3AI.get_instance().asubmit_impact(scope3_row)
    chat = ChatCompletionOutput(**asdict(response))
    chat.scope3ai = scope3ai_ctx
    return chat


# Todo: How headers works for stream
async def huggingface_async_chat_wrapper_stream(
    wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
) -> AsyncIterable[ChatCompletionStreamOutput]:
    timer_start = time.perf_counter()
    stream = await wrapped(*args, **kwargs)
    token_count = 0
    model = instance.model or kwargs["model"]
    async for chunk in stream:
        token_count += 1
        request_latency = time.perf_counter() - timer_start
        scope3_row = ImpactRow(
            model_id=model,
            output_tokens=token_count,
            request_duration_ms=request_latency
            * 1000,  # TODO: can we get the header that has the processing time
        )
        scope3_ctx = await Scope3AI.get_instance().asubmit_impact(scope3_row)
        chunk_data = ChatCompletionStreamOutput(**asdict(chunk))
        if scope3_ctx is not None:
            chunk_data.scope3ai = scope3_ctx
        yield chunk_data

1	import time	4✔
2	from collections.abc import AsyncIterable, Iterable	4✔
3	from dataclasses import asdict, dataclass	4✔
4	from typing import Any, Callable, Optional, Union	4✔
5
6	import tiktoken	4✔
7	from huggingface_hub import ( # type: ignore[import-untyped]	4✔
8	AsyncInferenceClient,
9	InferenceClient,
10	)
11	from huggingface_hub import ChatCompletionOutput as _ChatCompletionOutput	4✔
12	from huggingface_hub import ChatCompletionStreamOutput as _ChatCompletionStreamOutput	4✔
13	from requests import Response	4✔
14
15	from scope3ai.api.types import ImpactRow, Scope3AIContext	4✔
16	from scope3ai.lib import Scope3AI	4✔
17	from scope3ai.response_interceptor.requests_interceptor import requests_response_capture	4✔
18
19	HUGGING_FACE_CHAT_TASK = "chat"	4✔
20
21
22	@dataclass	4✔
23	class ChatCompletionOutput(_ChatCompletionOutput):	4✔
24	scope3ai: Optional[Scope3AIContext] = None	4✔
25
26
27	@dataclass	4✔
28	class ChatCompletionStreamOutput(_ChatCompletionStreamOutput):	4✔
29	scope3ai: Optional[Scope3AIContext] = None	4✔
30
31
32	def huggingface_chat_wrapper(	4✔
33	wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
34	) -> Union[ChatCompletionOutput, Iterable[ChatCompletionStreamOutput]]:
35	if kwargs.get("stream", False):	4✔
36	return huggingface_chat_wrapper_stream(wrapped, instance, args, kwargs)	4✔
37	else:
38	return huggingface_chat_wrapper_non_stream(wrapped, instance, args, kwargs)	4✔
39
40
41	def huggingface_chat_wrapper_non_stream(	4✔
42	wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
43	) -> ChatCompletionOutput:
44	timer_start = time.perf_counter()	4✔
45	http_response: Response \| None = None	4✔
46	with requests_response_capture() as responses:	4✔
47	response = wrapped(args, *kwargs)	4✔
48	http_responses = responses.get()	4✔
49	if http_responses:	4✔
50	http_response = http_responses[0]	4✔
51	model = (	4✔
52	instance.model
53	or kwargs.get("model")
54	or instance.get_recommended_model(HUGGING_FACE_CHAT_TASK)
55	)
56	if http_response:	4✔
57	compute_time = http_response.headers.get("x-compute-time")	4✔
58	else:
59	compute_time = time.perf_counter() - timer_start	×
60	scope3_row = ImpactRow(	4✔
61	model_id=model,
62	input_tokens=response.usage.prompt_tokens,
63	output_tokens=response.usage.completion_tokens,
64	request_duration_ms=float(compute_time) * 1000,
65	)
66	scope3ai_ctx = Scope3AI.get_instance().submit_impact(scope3_row)	4✔
67	chat = ChatCompletionOutput(**asdict(response))	4✔
68	chat.scope3ai = scope3ai_ctx	4✔
69	return chat	4✔
70
71
72	def huggingface_chat_wrapper_stream(	4✔
73	wrapped: Callable, instance: InferenceClient, args: Any, kwargs: Any
74	) -> Iterable[ChatCompletionStreamOutput]:
75	timer_start = time.perf_counter()	4✔
76	stream = wrapped(args, *kwargs)	4✔
77	token_count = 0	4✔
78	model = (	4✔
79	instance.model
80	or kwargs.get("model")
81	or instance.get_recommended_model(HUGGING_FACE_CHAT_TASK)
82	)
83	for chunk in stream:	4✔
84	token_count += 1	4✔
85	request_latency = time.perf_counter() - timer_start	4✔
86	scope3_row = ImpactRow(	4✔
87	model_id=model,
88	output_tokens=token_count,
89	request_duration_ms=request_latency * 1000,
90	)
91	chunk_data = ChatCompletionStreamOutput(**asdict(chunk))	4✔
92	scope3_ctx = Scope3AI.get_instance().submit_impact(scope3_row)	4✔
93	if scope3_ctx is not None:	4✔
94	chunk_data.scope3ai = scope3_ctx	4✔
95	yield chunk_data	4✔
96
97
98	async def huggingface_async_chat_wrapper(	4✔
99	wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
100	) -> Union[ChatCompletionOutput, AsyncIterable[ChatCompletionStreamOutput]]:
101	if kwargs.get("stream", False):	4✔
102	return huggingface_async_chat_wrapper_stream(wrapped, instance, args, kwargs)	4✔
103	else:
104	return await huggingface_async_chat_wrapper_non_stream(	4✔
105	wrapped, instance, args, kwargs
106	)
107
108
109	async def huggingface_async_chat_wrapper_non_stream(	4✔
110	wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
111	) -> ChatCompletionOutput:
112	timer_start = time.perf_counter()	4✔
113
114	response = await wrapped(args, *kwargs)	4✔
115	request_latency = time.perf_counter() - timer_start	4✔
116	model = (	4✔
117	instance.model or kwargs.get("model") or instance.get_recommended_model("chat")
118	)
119	encoder = tiktoken.get_encoding("cl100k_base")	4✔
120	output_tokens = len(encoder.encode(response.choices[0].message.content))	4✔
121	scope3_row = ImpactRow(	4✔
122	model_id=model,
123	input_tokens=response.usage.prompt_tokens,
124	output_tokens=output_tokens,
125	request_duration_ms=request_latency * 1000,
126	)
127
128	scope3ai_ctx = await Scope3AI.get_instance().asubmit_impact(scope3_row)	4✔
129	chat = ChatCompletionOutput(**asdict(response))	4✔
130	chat.scope3ai = scope3ai_ctx	4✔
131	return chat	4✔
132
133
134	# Todo: How headers works for stream
135	async def huggingface_async_chat_wrapper_stream(	4✔
136	wrapped: Callable, instance: AsyncInferenceClient, args: Any, kwargs: Any
137	) -> AsyncIterable[ChatCompletionStreamOutput]:
138	timer_start = time.perf_counter()	4✔
139	stream = await wrapped(args, *kwargs)	4✔
140	token_count = 0	4✔
141	model = instance.model or kwargs["model"]	4✔
142	async for chunk in stream:	4✔
143	token_count += 1	4✔
144	request_latency = time.perf_counter() - timer_start	4✔
145	scope3_row = ImpactRow(	4✔
146	model_id=model,
147	output_tokens=token_count,
148	request_duration_ms=request_latency
149	* 1000, # TODO: can we get the header that has the processing time
150	)
151	scope3_ctx = await Scope3AI.get_instance().asubmit_impact(scope3_row)	4✔
152	chunk_data = ChatCompletionStreamOutput(**asdict(chunk))	4✔
153	if scope3_ctx is not None:	4✔
154	chunk_data.scope3ai = scope3_ctx	4✔
155	yield chunk_data	4✔

scope3data / scope3ai-py / 13396554446

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