pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2025-12-08 20:42:30 +08:00
Code Issues Packages Projects Releases Wiki Activity

Test chat API and refine ppt chunker (#42)

Browse Source
This commit is contained in:
KevinHuSh
2024-01-23 19:45:36 +08:00
committed by GitHub
parent 34b2ab3b2f
commit e32ef75e99
10 changed files with 226 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
rag/llm/chat_model.py
View File

@ -37,7 +37,7 @@ class GptTurbo(Base):
model=self.model_name,
messages=history,
**gen_conf)
return res.choices[0].message.content.strip()
return res.choices[0].message.content.strip(), res.usage.completion_tokens
from dashscope import Generation
@ -56,5 +56,5 @@ class QWenChat(Base):
result_format='message'
)
if response.status_code == HTTPStatus.OK:
return response.output.choices[0]['message']['content']
return response.message
return response.output.choices[0]['message']['content'], response.usage.output_tokens
return response.message, 0

6
rag/llm/cv_model.py
View File

@ -72,7 +72,7 @@ class GptV4(Base):
messages=self.prompt(b64),
max_tokens=max_tokens,
)
return res.choices[0].message.content.strip()
return res.choices[0].message.content.strip(), res.usage.total_tokens
class QWenCV(Base):
@ -87,5 +87,5 @@ class QWenCV(Base):
response = MultiModalConversation.call(model=self.model_name,
messages=self.prompt(self.image2base64(image)))
if response.status_code == HTTPStatus.OK:
return response.output.choices[0]['message']['content']
return response.message
return response.output.choices[0]['message']['content'], response.usage.output_tokens
return response.message, 0

40
rag/llm/embedding_model.py
View File

@ -36,6 +36,9 @@ class Base(ABC):
def encode(self, texts: list, batch_size=32):
raise NotImplementedError("Please implement encode method!")
def encode_queries(self, text: str):
raise NotImplementedError("Please implement encode method!")
class HuEmbedding(Base):
def __init__(self, key="", model_name=""):
@ -68,15 +71,18 @@ class HuEmbedding(Base):
class OpenAIEmbed(Base):
def __init__(self, key, model_name="text-embedding-ada-002"):
self.client = OpenAI(key)
self.client = OpenAI(api_key=key)
self.model_name = model_name
def encode(self, texts: list, batch_size=32):
token_count = 0
for t in texts: token_count += num_tokens_from_string(t)
res = self.client.embeddings.create(input=texts,
model=self.model_name)
return [d["embedding"] for d in res["data"]], token_count
return np.array([d.embedding for d in res.data]), res.usage.total_tokens
def encode_queries(self, text):
res = self.client.embeddings.create(input=[text],
model=self.model_name)
return np.array(res.data[0].embedding), res.usage.total_tokens
class QWenEmbed(Base):
@ -84,16 +90,28 @@ class QWenEmbed(Base):
dashscope.api_key = key
self.model_name = model_name
def encode(self, texts: list, batch_size=32, text_type="document"):
def encode(self, texts: list, batch_size=10):
import dashscope
res = []
token_count = 0
for txt in texts:
texts = [txt[:2048] for txt in texts]
for i in range(0, len(texts), batch_size):
resp = dashscope.TextEmbedding.call(
model=self.model_name,
input=txt[:2048],
text_type=text_type
input=texts[i:i+batch_size],
text_type="document"
)
res.append(resp["output"]["embeddings"][0]["embedding"])
token_count += resp["usage"]["total_tokens"]
return res, token_count
embds = [[]] * len(resp["output"]["embeddings"])
for e in resp["output"]["embeddings"]:
embds[e["text_index"]] = e["embedding"]
res.extend(embds)
token_count += resp["usage"]["input_tokens"]
return np.array(res), token_count
def encode_queries(self, text):
resp = dashscope.TextEmbedding.call(
model=self.model_name,
input=text[:2048],
text_type="query"
)
return np.array(resp["output"]["embeddings"][0]["embedding"]), resp["usage"]["input_tokens"]