github-starred/open-webui
2
0
Fork 0
mirror of https://github.com/open-webui/open-webui.git synced 2026-05-01 17:59:28 -05:00
Code Issues 915 Packages Projects Releases 126 Wiki Activity

refac

Browse Source
This commit is contained in:
Timothy Jaeryang Baek
2026-03-17 17:58:01 -05:00
parent fcf7208352
commit de3317e26b
220 changed files with 17200 additions and 22836 deletions
Download Patch File Download Diff File Expand all files Collapse all files

174
backend/open_webui/utils/response.py
View File

@@ -21,28 +21,28 @@ def normalize_usage(usage: dict) -> dict:
# Map various field names to standard names
input_tokens = (
usage.get("input_tokens") # Already standard
or usage.get("prompt_tokens") # OpenAI
or usage.get("prompt_eval_count") # Ollama
or usage.get("prompt_n") # llama.cpp
usage.get('input_tokens') # Already standard
or usage.get('prompt_tokens') # OpenAI
or usage.get('prompt_eval_count') # Ollama
or usage.get('prompt_n') # llama.cpp
or 0
)
output_tokens = (
usage.get("output_tokens") # Already standard
or usage.get("completion_tokens") # OpenAI
or usage.get("eval_count") # Ollama
or usage.get("predicted_n") # llama.cpp
usage.get('output_tokens') # Already standard
or usage.get('completion_tokens') # OpenAI
or usage.get('eval_count') # Ollama
or usage.get('predicted_n') # llama.cpp
or 0
)
total_tokens = usage.get("total_tokens") or (input_tokens + output_tokens)
total_tokens = usage.get('total_tokens') or (input_tokens + output_tokens)
# Add standardized fields to original data
result = dict(usage)
result["input_tokens"] = int(input_tokens)
result["output_tokens"] = int(output_tokens)
result["total_tokens"] = int(total_tokens)
result['input_tokens'] = int(input_tokens)
result['output_tokens'] = int(output_tokens)
result['total_tokens'] = int(total_tokens)
return result
@@ -50,14 +50,14 @@ def normalize_usage(usage: dict) -> dict:
def convert_ollama_tool_call_to_openai(tool_calls: list) -> list:
openai_tool_calls = []
for tool_call in tool_calls:
function = tool_call.get("function", {})
function = tool_call.get('function', {})
openai_tool_call = {
"index": tool_call.get("index", function.get("index", 0)),
"id": tool_call.get("id", f"call_{str(uuid4())}"),
"type": "function",
"function": {
"name": function.get("name", ""),
"arguments": json.dumps(function.get("arguments", {})),
'index': tool_call.get('index', function.get('index', 0)),
'id': tool_call.get('id', f'call_{str(uuid4())}'),
'type': 'function',
'function': {
'name': function.get('name', ''),
'arguments': json.dumps(function.get('arguments', {})),
},
}
openai_tool_calls.append(openai_tool_call)
@@ -65,69 +65,57 @@ def convert_ollama_tool_call_to_openai(tool_calls: list) -> list:
def convert_ollama_usage_to_openai(data: dict) -> dict:
input_tokens = int(data.get("prompt_eval_count", 0))
output_tokens = int(data.get("eval_count", 0))
input_tokens = int(data.get('prompt_eval_count', 0))
output_tokens = int(data.get('eval_count', 0))
total_tokens = input_tokens + output_tokens
return {
# Standardized fields
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"total_tokens": total_tokens,
'input_tokens': input_tokens,
'output_tokens': output_tokens,
'total_tokens': total_tokens,
# OpenAI-compatible fields (for backward compatibility)
"prompt_tokens": input_tokens,
"completion_tokens": output_tokens,
'prompt_tokens': input_tokens,
'completion_tokens': output_tokens,
# Ollama-specific metrics
"response_token/s": (
'response_token/s': (
round(
(
(
data.get("eval_count", 0)
/ ((data.get("eval_duration", 0) / 10_000_000))
)
* 100
),
((data.get('eval_count', 0) / (data.get('eval_duration', 0) / 10_000_000)) * 100),
2,
)
if data.get("eval_duration", 0) > 0
else "N/A"
if data.get('eval_duration', 0) > 0
else 'N/A'
),
"prompt_token/s": (
'prompt_token/s': (
round(
(
(
data.get("prompt_eval_count", 0)
/ ((data.get("prompt_eval_duration", 0) / 10_000_000))
)
* 100
),
((data.get('prompt_eval_count', 0) / (data.get('prompt_eval_duration', 0) / 10_000_000)) * 100),
2,
)
if data.get("prompt_eval_duration", 0) > 0
else "N/A"
if data.get('prompt_eval_duration', 0) > 0
else 'N/A'
),
"total_duration": data.get("total_duration", 0),
"load_duration": data.get("load_duration", 0),
"prompt_eval_count": data.get("prompt_eval_count", 0),
"prompt_eval_duration": data.get("prompt_eval_duration", 0),
"eval_count": data.get("eval_count", 0),
"eval_duration": data.get("eval_duration", 0),
"approximate_total": (lambda s: f"{s // 3600}h{(s % 3600) // 60}m{s % 60}s")(
(data.get("total_duration", 0) or 0) // 1_000_000_000
'total_duration': data.get('total_duration', 0),
'load_duration': data.get('load_duration', 0),
'prompt_eval_count': data.get('prompt_eval_count', 0),
'prompt_eval_duration': data.get('prompt_eval_duration', 0),
'eval_count': data.get('eval_count', 0),
'eval_duration': data.get('eval_duration', 0),
'approximate_total': (lambda s: f'{s // 3600}h{(s % 3600) // 60}m{s % 60}s')(
(data.get('total_duration', 0) or 0) // 1_000_000_000
),
"completion_tokens_details": {
"reasoning_tokens": 0,
"accepted_prediction_tokens": 0,
"rejected_prediction_tokens": 0,
'completion_tokens_details': {
'reasoning_tokens': 0,
'accepted_prediction_tokens': 0,
'rejected_prediction_tokens': 0,
},
}
def convert_response_ollama_to_openai(ollama_response: dict) -> dict:
model = ollama_response.get("model", "ollama")
message_content = ollama_response.get("message", {}).get("content", "")
reasoning_content = ollama_response.get("message", {}).get("thinking", None)
tool_calls = ollama_response.get("message", {}).get("tool_calls", None)
model = ollama_response.get('model', 'ollama')
message_content = ollama_response.get('message', {}).get('content', '')
reasoning_content = ollama_response.get('message', {}).get('thinking', None)
tool_calls = ollama_response.get('message', {}).get('tool_calls', None)
openai_tool_calls = None
if tool_calls:
@@ -148,33 +136,31 @@ async def convert_streaming_response_ollama_to_openai(ollama_streaming_response)
async for data in ollama_streaming_response.body_iterator:
data = json.loads(data)
model = data.get("model", "ollama")
message_content = data.get("message", {}).get("content", None)
reasoning_content = data.get("message", {}).get("thinking", None)
tool_calls = data.get("message", {}).get("tool_calls", None)
model = data.get('model', 'ollama')
message_content = data.get('message', {}).get('content', None)
reasoning_content = data.get('message', {}).get('thinking', None)
tool_calls = data.get('message', {}).get('tool_calls', None)
openai_tool_calls = None
if tool_calls:
openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
has_tool_calls = True
done = data.get("done", False)
done = data.get('done', False)
usage = None
if done:
usage = convert_ollama_usage_to_openai(data)
data = openai_chat_chunk_message_template(
model, message_content, reasoning_content, openai_tool_calls, usage
)
data = openai_chat_chunk_message_template(model, message_content, reasoning_content, openai_tool_calls, usage)
if done and has_tool_calls:
data["choices"][0]["finish_reason"] = "tool_calls"
data['choices'][0]['finish_reason'] = 'tool_calls'
line = f"data: {json.dumps(data)}\n\n"
line = f'data: {json.dumps(data)}\n\n'
yield line
yield "data: [DONE]\n\n"
yield 'data: [DONE]\n\n'
def convert_embedding_response_ollama_to_openai(response) -> dict:
@@ -199,51 +185,47 @@ def convert_embedding_response_ollama_to_openai(response) -> dict:
"""
# Ollama batch-style output from /api/embed
# Response format: {"embeddings": [[0.1, 0.2, ...], [0.3, 0.4, ...]], "model": "..."}
if isinstance(response, dict) and "embeddings" in response:
if isinstance(response, dict) and 'embeddings' in response:
openai_data = []
for i, emb in enumerate(response["embeddings"]):
for i, emb in enumerate(response['embeddings']):
# /api/embed returns embeddings as plain float lists
if isinstance(emb, list):
openai_data.append(
{
"object": "embedding",
"embedding": emb,
"index": i,
'object': 'embedding',
'embedding': emb,
'index': i,
}
)
# Also handle dict format for robustness
elif isinstance(emb, dict):
openai_data.append(
{
"object": "embedding",
"embedding": emb.get("embedding"),
"index": emb.get("index", i),
'object': 'embedding',
'embedding': emb.get('embedding'),
'index': emb.get('index', i),
}
)
return {
"object": "list",
"data": openai_data,
"model": response.get("model"),
'object': 'list',
'data': openai_data,
'model': response.get('model'),
}
# Ollama single output
elif isinstance(response, dict) and "embedding" in response:
elif isinstance(response, dict) and 'embedding' in response:
return {
"object": "list",
"data": [
'object': 'list',
'data': [
{
"object": "embedding",
"embedding": response["embedding"],
"index": 0,
'object': 'embedding',
'embedding': response['embedding'],
'index': 0,
}
],
"model": response.get("model"),
'model': response.get('model'),
}
# Already OpenAI-compatible?
elif (
isinstance(response, dict)
and "data" in response
and isinstance(response["data"], list)
):
elif isinstance(response, dict) and 'data' in response and isinstance(response['data'], list):
return response
# Fallback: return as is if unrecognized