wehub-resource-sync
264 lines
8.6 KiB
Python
264 lines
8.6 KiB
Python
# Copyright 2026 Emcie Co Ltd.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
from __future__ import annotations
|
|
import time
|
|
from openai import (
|
|
APIConnectionError,
|
|
APIResponseValidationError,
|
|
APITimeoutError,
|
|
AsyncClient,
|
|
ConflictError,
|
|
InternalServerError,
|
|
RateLimitError,
|
|
)
|
|
from typing import Any, Mapping
|
|
from typing_extensions import override
|
|
import json
|
|
import jsonfinder # type: ignore
|
|
import os
|
|
|
|
from pydantic import ValidationError
|
|
import tiktoken
|
|
|
|
from parlant.adapters.nlp.common import normalize_json_output, record_llm_metrics
|
|
from parlant.adapters.nlp.hugging_face import JinaAIEmbedder
|
|
from parlant.core.engines.alpha.prompt_builder import PromptBuilder
|
|
from parlant.core.loggers import Logger
|
|
from parlant.core.tracer import Tracer
|
|
from parlant.core.meter import Meter
|
|
from parlant.core.nlp.policies import policy, retry
|
|
from parlant.core.nlp.tokenization import EstimatingTokenizer
|
|
from parlant.core.nlp.service import (
|
|
EmbedderHints,
|
|
NLPService,
|
|
SchematicGeneratorHints,
|
|
StreamingTextGeneratorHints,
|
|
)
|
|
from parlant.core.nlp.embedding import Embedder
|
|
from parlant.core.nlp.generation import (
|
|
T,
|
|
BaseSchematicGenerator,
|
|
SchematicGenerationResult,
|
|
StreamingTextGenerator,
|
|
)
|
|
from parlant.core.nlp.generation_info import GenerationInfo, UsageInfo
|
|
from parlant.core.nlp.moderation import (
|
|
ModerationService,
|
|
NoModeration,
|
|
)
|
|
from parlant.core.health import HealthReporter
|
|
|
|
|
|
class DeepSeekEstimatingTokenizer(EstimatingTokenizer):
|
|
def __init__(self, model_name: str) -> None:
|
|
self.model_name = model_name
|
|
self.encoding = tiktoken.encoding_for_model("gpt-4o-2024-08-06")
|
|
|
|
@override
|
|
async def estimate_token_count(self, prompt: str) -> int:
|
|
tokens = self.encoding.encode(prompt)
|
|
return len(tokens)
|
|
|
|
|
|
class DeepSeekSchematicGenerator(BaseSchematicGenerator[T]):
|
|
supported_deepseek_params = ["temperature", "logit_bias", "max_tokens"]
|
|
supported_hints = supported_deepseek_params + ["strict"]
|
|
|
|
def __init__(self,
|
|
model_name: str,
|
|
logger: Logger,
|
|
tracer: Tracer,
|
|
meter: Meter, health_reporter: HealthReporter,
|
|
) -> None:
|
|
super().__init__(logger=logger, tracer=tracer, meter=meter, health_reporter=health_reporter, model_name=model_name)
|
|
|
|
self._client = AsyncClient(
|
|
base_url="https://api.deepseek.com",
|
|
api_key=os.environ["DEEPSEEK_API_KEY"],
|
|
)
|
|
|
|
self._tokenizer = DeepSeekEstimatingTokenizer(model_name=self.model_name)
|
|
|
|
@property
|
|
@override
|
|
def id(self) -> str:
|
|
return f"deepseek/{self.model_name}"
|
|
|
|
@property
|
|
@override
|
|
def tokenizer(self) -> DeepSeekEstimatingTokenizer:
|
|
return self._tokenizer
|
|
|
|
@policy(
|
|
[
|
|
retry(
|
|
exceptions=(
|
|
APIConnectionError,
|
|
APITimeoutError,
|
|
ConflictError,
|
|
RateLimitError,
|
|
APIResponseValidationError,
|
|
),
|
|
),
|
|
retry(InternalServerError, max_exceptions=2, wait_times=(1.0, 5.0)),
|
|
]
|
|
)
|
|
@override
|
|
async def do_generate(
|
|
self,
|
|
prompt: str | PromptBuilder,
|
|
hints: Mapping[str, Any] = {},
|
|
) -> SchematicGenerationResult[T]:
|
|
with self.logger.scope(f"DeepSeek LLM Request ({self.schema.__name__})"):
|
|
return await self._do_generate(prompt, hints)
|
|
|
|
async def _do_generate(
|
|
self,
|
|
prompt: str | PromptBuilder,
|
|
hints: Mapping[str, Any] = {},
|
|
) -> SchematicGenerationResult[T]:
|
|
if isinstance(prompt, PromptBuilder):
|
|
prompt = prompt.build()
|
|
|
|
deepseek_api_arguments = {
|
|
k: v for k, v in hints.items() if k in self.supported_deepseek_params
|
|
}
|
|
|
|
t_start = time.time()
|
|
response = await self._client.chat.completions.create(
|
|
messages=[{"role": "user", "content": prompt}],
|
|
model=self.model_name,
|
|
max_tokens=8192,
|
|
response_format={"type": "json_object"},
|
|
**deepseek_api_arguments,
|
|
)
|
|
t_end = time.time()
|
|
|
|
if response.usage:
|
|
self.logger.trace(response.usage.model_dump_json(indent=2))
|
|
|
|
raw_content = response.choices[0].message.content or "{}"
|
|
|
|
try:
|
|
json_content = json.loads(normalize_json_output(raw_content))
|
|
except json.JSONDecodeError:
|
|
self.logger.warning(f"Invalid JSON returned by {self.model_name}:\n{raw_content})")
|
|
json_content = jsonfinder.only_json(raw_content)[2]
|
|
self.logger.warning("Found JSON content within model response; continuing...")
|
|
|
|
try:
|
|
content = self.schema.model_validate(json_content)
|
|
|
|
assert response.usage
|
|
|
|
await record_llm_metrics(
|
|
self.meter,
|
|
self.model_name,
|
|
schema_name=self.schema.__name__,
|
|
input_tokens=response.usage.prompt_tokens,
|
|
output_tokens=response.usage.completion_tokens,
|
|
cached_input_tokens=getattr(
|
|
response,
|
|
"usage.prompt_cache_hit_tokens",
|
|
0,
|
|
),
|
|
)
|
|
|
|
return SchematicGenerationResult(
|
|
content=content,
|
|
info=GenerationInfo(
|
|
schema_name=self.schema.__name__,
|
|
model=self.id,
|
|
duration=(t_end - t_start),
|
|
usage=UsageInfo(
|
|
input_tokens=response.usage.prompt_tokens,
|
|
output_tokens=response.usage.completion_tokens,
|
|
extra={
|
|
"cached_input_tokens": getattr(
|
|
response,
|
|
"usage.prompt_cache_hit_tokens",
|
|
0,
|
|
)
|
|
},
|
|
),
|
|
),
|
|
)
|
|
except ValidationError:
|
|
self.logger.error(
|
|
f"JSON content returned by {self.model_name} does not match expected schema:\n{raw_content}"
|
|
)
|
|
raise
|
|
|
|
|
|
class DeepSeek_Chat(DeepSeekSchematicGenerator[T]):
|
|
def __init__(self, logger: Logger, tracer: Tracer, meter: Meter, health_reporter: HealthReporter) -> None:
|
|
super().__init__(model_name="deepseek-chat", logger=logger, tracer=tracer, meter=meter, health_reporter=health_reporter)
|
|
|
|
@property
|
|
@override
|
|
def max_tokens(self) -> int:
|
|
return 128 * 1024
|
|
|
|
|
|
class DeepSeekService(NLPService):
|
|
@staticmethod
|
|
def verify_environment() -> str | None:
|
|
"""Returns an error message if the environment is not set up correctly."""
|
|
|
|
if not os.environ.get("DEEPSEEK_API_KEY"):
|
|
return """\
|
|
You're using the DeepSeek NLP service, but DEEPSEEK_API_KEY is not set.
|
|
Please set DEEPSEEK_API_KEY in your environment before running Parlant.
|
|
"""
|
|
|
|
return None
|
|
|
|
def __init__(self,
|
|
logger: Logger,
|
|
tracer: Tracer,
|
|
meter: Meter, health_reporter: HealthReporter,
|
|
) -> None:
|
|
self._logger = logger
|
|
self._tracer = tracer
|
|
self._meter = meter
|
|
|
|
self._health_reporter = health_reporter
|
|
self._logger.info("Initialized DeepSeekService")
|
|
|
|
@property
|
|
@override
|
|
def supports_streaming(self) -> bool:
|
|
return False
|
|
|
|
@override
|
|
async def get_streaming_text_generator(
|
|
self, hints: StreamingTextGeneratorHints = {}
|
|
) -> StreamingTextGenerator:
|
|
raise NotImplementedError("Streaming is not supported. Check supports_streaming first.")
|
|
|
|
@override
|
|
async def get_schematic_generator(
|
|
self, t: type[T], hints: SchematicGeneratorHints = {}
|
|
) -> DeepSeekSchematicGenerator[T]:
|
|
return DeepSeek_Chat[t](self._logger, self._tracer, self._meter, self._health_reporter) # type: ignore
|
|
|
|
@override
|
|
async def get_embedder(self, hints: EmbedderHints = {}) -> Embedder:
|
|
return JinaAIEmbedder(self._logger, self._tracer, self._meter, self._health_reporter)
|
|
|
|
@override
|
|
async def get_moderation_service(self) -> ModerationService:
|
|
return NoModeration()
|