Spaces:
Build error
Build error
File size: 8,211 Bytes
d6703a1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 |
import json
from uuid import uuid4
from open_webui.utils.misc import (
openai_chat_chunk_message_template,
openai_chat_completion_message_template,
)
def normalize_usage(usage: dict) -> dict:
"""
Normalize usage statistics to standard format.
Handles OpenAI, Ollama, and llama.cpp formats.
Adds standardized token fields to the original data:
- input_tokens: Number of tokens in the prompt
- output_tokens: Number of tokens generated
- total_tokens: Sum of input and output tokens
"""
if not usage:
return {}
# 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
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
or 0
)
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)
return result
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", {})
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", {})),
},
}
openai_tool_calls.append(openai_tool_call)
return openai_tool_calls
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))
total_tokens = input_tokens + output_tokens
return {
# Standardized fields
"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,
# Ollama-specific metrics
"response_token/s": (
round(
(
(
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"
),
"prompt_token/s": (
round(
(
(
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"
),
"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,
},
}
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)
openai_tool_calls = None
if tool_calls:
openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
data = ollama_response
usage = convert_ollama_usage_to_openai(data)
response = openai_chat_completion_message_template(
model, message_content, reasoning_content, openai_tool_calls, usage
)
return response
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)
openai_tool_calls = None
if tool_calls:
openai_tool_calls = convert_ollama_tool_call_to_openai(tool_calls)
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
)
line = f"data: {json.dumps(data)}\n\n"
yield line
yield "data: [DONE]\n\n"
def convert_embedding_response_ollama_to_openai(response) -> dict:
"""
Convert the response from Ollama embeddings endpoint to the OpenAI-compatible format.
Args:
response (dict): The response from the Ollama API,
e.g. {"embedding": [...], "model": "..."}
or {"embeddings": [{"embedding": [...], "index": 0}, ...], "model": "..."}
Returns:
dict: Response adapted to OpenAI's embeddings API format.
e.g. {
"object": "list",
"data": [
{"object": "embedding", "embedding": [...], "index": 0},
...
],
"model": "...",
}
"""
# 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:
openai_data = []
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,
}
)
# Also handle dict format for robustness
elif isinstance(emb, dict):
openai_data.append(
{
"object": "embedding",
"embedding": emb.get("embedding"),
"index": emb.get("index", i),
}
)
return {
"object": "list",
"data": openai_data,
"model": response.get("model"),
}
# Ollama single output
elif isinstance(response, dict) and "embedding" in response:
return {
"object": "list",
"data": [
{
"object": "embedding",
"embedding": response["embedding"],
"index": 0,
}
],
"model": response.get("model"),
}
# Already OpenAI-compatible?
elif (
isinstance(response, dict)
and "data" in response
and isinstance(response["data"], list)
):
return response
# Fallback: return as is if unrecognized
return response
|