openlit 0.0.1__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- openlit/__helpers.py +143 -0
- openlit/__init__.py +184 -0
- openlit/instrumentation/anthropic/__init__.py +50 -0
- openlit/instrumentation/anthropic/anthropic.py +291 -0
- openlit/instrumentation/anthropic/async_anthropic.py +291 -0
- openlit/instrumentation/chroma/__init__.py +86 -0
- openlit/instrumentation/chroma/chroma.py +197 -0
- openlit/instrumentation/cohere/__init__.py +51 -0
- openlit/instrumentation/cohere/cohere.py +397 -0
- openlit/instrumentation/langchain/__init__.py +74 -0
- openlit/instrumentation/langchain/langchain.py +161 -0
- openlit/instrumentation/mistral/__init__.py +80 -0
- openlit/instrumentation/mistral/async_mistral.py +417 -0
- openlit/instrumentation/mistral/mistral.py +416 -0
- openlit/instrumentation/openai/__init__.py +335 -0
- openlit/instrumentation/openai/async_azure_openai.py +841 -0
- openlit/instrumentation/openai/async_openai.py +875 -0
- openlit/instrumentation/openai/azure_openai.py +840 -0
- openlit/instrumentation/openai/openai.py +891 -0
- openlit/instrumentation/pinecone/__init__.py +66 -0
- openlit/instrumentation/pinecone/pinecone.py +173 -0
- openlit/instrumentation/transformers/__init__.py +37 -0
- openlit/instrumentation/transformers/transformers.py +156 -0
- openlit/otel/metrics.py +109 -0
- openlit/otel/tracing.py +83 -0
- openlit/semcov/__init__.py +123 -0
- openlit-0.0.1.dist-info/LICENSE +201 -0
- openlit-0.0.1.dist-info/METADATA +113 -0
- openlit-0.0.1.dist-info/RECORD +30 -0
- openlit-0.0.1.dist-info/WHEEL +4 -0
@@ -0,0 +1,397 @@
|
|
1
|
+
# pylint: disable=duplicate-code, broad-exception-caught, too-many-statements, unused-argument
|
2
|
+
"""
|
3
|
+
Module for monitoring Cohere API calls.
|
4
|
+
"""
|
5
|
+
|
6
|
+
import logging
|
7
|
+
from opentelemetry.trace import SpanKind, Status, StatusCode
|
8
|
+
from opentelemetry.sdk.resources import TELEMETRY_SDK_NAME
|
9
|
+
from openlit.__helpers import get_chat_model_cost, get_embed_model_cost, handle_exception
|
10
|
+
from openlit.semcov import SemanticConvetion
|
11
|
+
|
12
|
+
# Initialize logger for logging potential issues and operations
|
13
|
+
logger = logging.getLogger(__name__)
|
14
|
+
|
15
|
+
def embed(gen_ai_endpoint, version, environment, application_name, tracer,
|
16
|
+
pricing_info, trace_content, metrics, disable_metrics):
|
17
|
+
"""
|
18
|
+
Generates a telemetry wrapper for embeddings to collect metrics.
|
19
|
+
|
20
|
+
Args:
|
21
|
+
gen_ai_endpoint: Endpoint identifier for logging and tracing.
|
22
|
+
version: Version of the monitoring package.
|
23
|
+
environment: Deployment environment (e.g., production, staging).
|
24
|
+
application_name: Name of the application using the OpenAI API.
|
25
|
+
tracer: OpenTelemetry tracer for creating spans.
|
26
|
+
pricing_info: Information used for calculating the cost of OpenAI usage.
|
27
|
+
trace_content: Flag indicating whether to trace the actual content.
|
28
|
+
|
29
|
+
Returns:
|
30
|
+
A function that wraps the embeddings method to add telemetry.
|
31
|
+
"""
|
32
|
+
|
33
|
+
def wrapper(wrapped, instance, args, kwargs):
|
34
|
+
"""
|
35
|
+
Wraps the 'embed' API call to add telemetry.
|
36
|
+
|
37
|
+
This collects metrics such as execution time, cost, and token usage, and handles errors
|
38
|
+
gracefully, adding details to the trace for observability.
|
39
|
+
|
40
|
+
Args:
|
41
|
+
wrapped: The original 'embed' method to be wrapped.
|
42
|
+
instance: The instance of the class where the original method is defined.
|
43
|
+
args: Positional arguments for the 'embed' method.
|
44
|
+
kwargs: Keyword arguments for the 'embed' method.
|
45
|
+
|
46
|
+
Returns:
|
47
|
+
The response from the original 'embed' method.
|
48
|
+
"""
|
49
|
+
|
50
|
+
with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
|
51
|
+
response = wrapped(*args, **kwargs)
|
52
|
+
|
53
|
+
try:
|
54
|
+
# Get prompt from kwargs and store as a single string
|
55
|
+
prompt = " ".join(kwargs.get("texts", []))
|
56
|
+
|
57
|
+
|
58
|
+
# Calculate cost of the operation
|
59
|
+
cost = get_embed_model_cost(kwargs.get("model", "embed-english-v2.0"),
|
60
|
+
pricing_info,
|
61
|
+
response.meta.billed_units.input_tokens)
|
62
|
+
|
63
|
+
# Set Span attributes
|
64
|
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
|
65
|
+
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
|
66
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE)
|
67
|
+
span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
|
68
|
+
SemanticConvetion.GEN_AI_TYPE_EMBEDDING)
|
69
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
|
70
|
+
gen_ai_endpoint)
|
71
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
|
72
|
+
environment)
|
73
|
+
span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
|
74
|
+
application_name)
|
75
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
|
76
|
+
kwargs.get("model", "embed-english-v2.0"))
|
77
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_EMBEDDING_FORMAT,
|
78
|
+
kwargs.get("embedding_types", "float"))
|
79
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_EMBEDDING_DIMENSION,
|
80
|
+
kwargs.get("input_type", ""))
|
81
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_USER,
|
82
|
+
kwargs.get("user", ""))
|
83
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_ID,
|
84
|
+
response.id)
|
85
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
|
86
|
+
response.meta.billed_units.input_tokens)
|
87
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
|
88
|
+
response.meta.billed_units.input_tokens)
|
89
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
|
90
|
+
cost)
|
91
|
+
if trace_content:
|
92
|
+
span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_PROMPT,
|
93
|
+
prompt)
|
94
|
+
|
95
|
+
span.set_status(Status(StatusCode.OK))
|
96
|
+
|
97
|
+
if disable_metrics is False:
|
98
|
+
attributes = {
|
99
|
+
TELEMETRY_SDK_NAME:
|
100
|
+
"openlit",
|
101
|
+
SemanticConvetion.GEN_AI_APPLICATION_NAME:
|
102
|
+
application_name,
|
103
|
+
SemanticConvetion.GEN_AI_SYSTEM:
|
104
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE,
|
105
|
+
SemanticConvetion.GEN_AI_ENVIRONMENT:
|
106
|
+
environment,
|
107
|
+
SemanticConvetion.GEN_AI_TYPE:
|
108
|
+
SemanticConvetion.GEN_AI_TYPE_EMBEDDING,
|
109
|
+
SemanticConvetion.GEN_AI_REQUEST_MODEL:
|
110
|
+
kwargs.get("model", "embed-english-v2.0")
|
111
|
+
}
|
112
|
+
|
113
|
+
metrics["genai_requests"].add(1, attributes)
|
114
|
+
metrics["genai_total_tokens"].add(
|
115
|
+
response.meta.billed_units.input_tokens, attributes
|
116
|
+
)
|
117
|
+
metrics["genai_prompt_tokens"].add(
|
118
|
+
response.meta.billed_units.input_tokens, attributes
|
119
|
+
)
|
120
|
+
metrics["genai_cost"].record(cost, attributes)
|
121
|
+
|
122
|
+
# Return original response
|
123
|
+
return response
|
124
|
+
|
125
|
+
except Exception as e:
|
126
|
+
handle_exception(span, e)
|
127
|
+
logger.error("Error in trace creation: %s", e)
|
128
|
+
|
129
|
+
# Return original response
|
130
|
+
return response
|
131
|
+
|
132
|
+
return wrapper
|
133
|
+
|
134
|
+
def chat(gen_ai_endpoint, version, environment, application_name, tracer,
|
135
|
+
pricing_info, trace_content, metrics, disable_metrics):
|
136
|
+
"""
|
137
|
+
Generates a telemetry wrapper for chat to collect metrics.
|
138
|
+
|
139
|
+
Args:
|
140
|
+
gen_ai_endpoint: Endpoint identifier for logging and tracing.
|
141
|
+
version: Version of the monitoring package.
|
142
|
+
environment: Deployment environment (e.g., production, staging).
|
143
|
+
application_name: Name of the application using the OpenAI API.
|
144
|
+
tracer: OpenTelemetry tracer for creating spans.
|
145
|
+
pricing_info: Information used for calculating the cost of OpenAI usage.
|
146
|
+
trace_content: Flag indicating whether to trace the actual content.
|
147
|
+
|
148
|
+
Returns:
|
149
|
+
A function that wraps the chat method to add telemetry.
|
150
|
+
"""
|
151
|
+
|
152
|
+
def wrapper(wrapped, instance, args, kwargs):
|
153
|
+
"""
|
154
|
+
Wraps the 'chat' API call to add telemetry.
|
155
|
+
|
156
|
+
This collects metrics such as execution time, cost, and token usage, and handles errors
|
157
|
+
gracefully, adding details to the trace for observability.
|
158
|
+
|
159
|
+
Args:
|
160
|
+
wrapped: The original 'chat' method to be wrapped.
|
161
|
+
instance: The instance of the class where the original method is defined.
|
162
|
+
args: Positional arguments for the 'chat' method.
|
163
|
+
kwargs: Keyword arguments for the 'chat' method.
|
164
|
+
|
165
|
+
Returns:
|
166
|
+
The response from the original 'chat' method.
|
167
|
+
"""
|
168
|
+
|
169
|
+
with tracer.start_as_current_span(gen_ai_endpoint, kind=SpanKind.CLIENT) as span:
|
170
|
+
response = wrapped(*args, **kwargs)
|
171
|
+
|
172
|
+
try:
|
173
|
+
# Calculate cost of the operation
|
174
|
+
cost = get_chat_model_cost(kwargs.get("model", "command"),
|
175
|
+
pricing_info,
|
176
|
+
response.meta["billed_units"]["input_tokens"],
|
177
|
+
response.meta["billed_units"]["output_tokens"])
|
178
|
+
|
179
|
+
# Set Span attributes
|
180
|
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
|
181
|
+
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
|
182
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE)
|
183
|
+
span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
|
184
|
+
SemanticConvetion.GEN_AI_TYPE_CHAT)
|
185
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
|
186
|
+
gen_ai_endpoint)
|
187
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
|
188
|
+
environment)
|
189
|
+
span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
|
190
|
+
application_name)
|
191
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
|
192
|
+
kwargs.get("model", "command"))
|
193
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE,
|
194
|
+
kwargs.get("temperature", 0.3))
|
195
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MAX_TOKENS,
|
196
|
+
kwargs.get("max_tokens", ""))
|
197
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_SEED,
|
198
|
+
kwargs.get("seed", ""))
|
199
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_FREQUENCY_PENALTY,
|
200
|
+
kwargs.get("frequency_penalty", 0.0))
|
201
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_PRESENCE_PENALTY,
|
202
|
+
kwargs.get("presence_penalty", 0.0))
|
203
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
|
204
|
+
False)
|
205
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_ID,
|
206
|
+
response.response_id)
|
207
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_FINISH_REASON,
|
208
|
+
response.response_id)
|
209
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
|
210
|
+
response.meta["billed_units"]["input_tokens"])
|
211
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
|
212
|
+
response.meta["billed_units"]["output_tokens"])
|
213
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
|
214
|
+
response.meta["billed_units"]["input_tokens"] +
|
215
|
+
response.meta["billed_units"]["output_tokens"])
|
216
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
|
217
|
+
cost)
|
218
|
+
if trace_content:
|
219
|
+
span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_PROMPT,
|
220
|
+
kwargs.get("message", ""))
|
221
|
+
span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_COMPLETION,
|
222
|
+
response.text)
|
223
|
+
|
224
|
+
span.set_status(Status(StatusCode.OK))
|
225
|
+
|
226
|
+
if disable_metrics is False:
|
227
|
+
attributes = {
|
228
|
+
TELEMETRY_SDK_NAME:
|
229
|
+
"openlit",
|
230
|
+
SemanticConvetion.GEN_AI_APPLICATION_NAME:
|
231
|
+
application_name,
|
232
|
+
SemanticConvetion.GEN_AI_SYSTEM:
|
233
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE,
|
234
|
+
SemanticConvetion.GEN_AI_ENVIRONMENT:
|
235
|
+
environment,
|
236
|
+
SemanticConvetion.GEN_AI_TYPE:
|
237
|
+
SemanticConvetion.GEN_AI_TYPE_CHAT,
|
238
|
+
SemanticConvetion.GEN_AI_REQUEST_MODEL:
|
239
|
+
kwargs.get("model", "command")
|
240
|
+
}
|
241
|
+
|
242
|
+
metrics["genai_requests"].add(1, attributes)
|
243
|
+
metrics["genai_total_tokens"].add(
|
244
|
+
response.meta["billed_units"]["input_tokens"] +
|
245
|
+
response.meta["billed_units"]["output_tokens"], attributes)
|
246
|
+
metrics["genai_completion_tokens"].add(
|
247
|
+
response.meta["billed_units"]["output_tokens"], attributes)
|
248
|
+
metrics["genai_prompt_tokens"].add(
|
249
|
+
response.meta["billed_units"]["input_tokens"], attributes)
|
250
|
+
metrics["genai_cost"].record(cost, attributes)
|
251
|
+
|
252
|
+
# Return original response
|
253
|
+
return response
|
254
|
+
|
255
|
+
except Exception as e:
|
256
|
+
handle_exception(span, e)
|
257
|
+
logger.error("Error in trace creation: %s", e)
|
258
|
+
|
259
|
+
# Return original response
|
260
|
+
return response
|
261
|
+
|
262
|
+
return wrapper
|
263
|
+
|
264
|
+
def chat_stream(gen_ai_endpoint, version, environment, application_name,
|
265
|
+
tracer, pricing_info, trace_content, metrics, disable_metrics):
|
266
|
+
"""
|
267
|
+
Generates a telemetry wrapper for chat_stream to collect metrics.
|
268
|
+
|
269
|
+
Args:
|
270
|
+
gen_ai_endpoint: Endpoint identifier for logging and tracing.
|
271
|
+
version: Version of the monitoring package.
|
272
|
+
environment: Deployment environment (e.g., production, staging).
|
273
|
+
application_name: Name of the application using the OpenAI API.
|
274
|
+
tracer: OpenTelemetry tracer for creating spans.
|
275
|
+
pricing_info: Information used for calculating the cost of OpenAI usage.
|
276
|
+
trace_content: Flag indicating whether to trace the actual content.
|
277
|
+
|
278
|
+
Returns:
|
279
|
+
A function that wraps the chat method to add telemetry.
|
280
|
+
"""
|
281
|
+
|
282
|
+
def wrapper(wrapped, instance, args, kwargs):
|
283
|
+
"""
|
284
|
+
Wraps the 'chat_stream' API call to add telemetry.
|
285
|
+
|
286
|
+
This collects metrics such as execution time, cost, and token usage, and handles errors
|
287
|
+
gracefully, adding details to the trace for observability.
|
288
|
+
|
289
|
+
Args:
|
290
|
+
wrapped: The original 'chat_stream' method to be wrapped.
|
291
|
+
instance: The instance of the class where the original method is defined.
|
292
|
+
args: Positional arguments for the 'chat_stream' method.
|
293
|
+
kwargs: Keyword arguments for the 'chat_stream' method.
|
294
|
+
|
295
|
+
Returns:
|
296
|
+
The response from the original 'chat_stream' method.
|
297
|
+
"""
|
298
|
+
|
299
|
+
def stream_generator():
|
300
|
+
with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
|
301
|
+
# Placeholder for aggregating streaming response
|
302
|
+
llmresponse = ""
|
303
|
+
|
304
|
+
# Loop through streaming events capturing relevant details
|
305
|
+
for event in wrapped(*args, **kwargs):
|
306
|
+
# Collect message IDs and aggregated response from events
|
307
|
+
if event.event_type == "stream-end":
|
308
|
+
llmresponse = event.response.text
|
309
|
+
response_id = event.response.response_id
|
310
|
+
prompt_tokens = event.response.meta["billed_units"]["input_tokens"]
|
311
|
+
completion_tokens = event.response.meta["billed_units"]["output_tokens"]
|
312
|
+
finish_reason = event.finish_reason
|
313
|
+
yield event
|
314
|
+
|
315
|
+
# Handling exception ensure observability without disrupting operation
|
316
|
+
try:
|
317
|
+
# Calculate cost of the operation
|
318
|
+
cost = get_chat_model_cost(kwargs.get("model", "command"),
|
319
|
+
pricing_info, prompt_tokens, completion_tokens)
|
320
|
+
|
321
|
+
# Set Span attributes
|
322
|
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
|
323
|
+
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
|
324
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE)
|
325
|
+
span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
|
326
|
+
SemanticConvetion.GEN_AI_TYPE_CHAT)
|
327
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
|
328
|
+
gen_ai_endpoint)
|
329
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
|
330
|
+
environment)
|
331
|
+
span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
|
332
|
+
application_name)
|
333
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
|
334
|
+
kwargs.get("model", "command"))
|
335
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE,
|
336
|
+
kwargs.get("temperature", 0.3))
|
337
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MAX_TOKENS,
|
338
|
+
kwargs.get("max_tokens", ""))
|
339
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_SEED,
|
340
|
+
kwargs.get("seed", ""))
|
341
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_FREQUENCY_PENALTY,
|
342
|
+
kwargs.get("frequency_penalty", 0.0))
|
343
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_PRESENCE_PENALTY,
|
344
|
+
kwargs.get("presence_penalty", 0.0))
|
345
|
+
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
|
346
|
+
True)
|
347
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_ID,
|
348
|
+
response_id)
|
349
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_FINISH_REASON,
|
350
|
+
finish_reason)
|
351
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
|
352
|
+
prompt_tokens)
|
353
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
|
354
|
+
completion_tokens)
|
355
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
|
356
|
+
prompt_tokens + completion_tokens)
|
357
|
+
span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
|
358
|
+
cost)
|
359
|
+
if trace_content:
|
360
|
+
span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_PROMPT,
|
361
|
+
kwargs.get("message", ""))
|
362
|
+
span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_COMPLETION,
|
363
|
+
llmresponse)
|
364
|
+
|
365
|
+
span.set_status(Status(StatusCode.OK))
|
366
|
+
|
367
|
+
if disable_metrics is False:
|
368
|
+
attributes = {
|
369
|
+
TELEMETRY_SDK_NAME:
|
370
|
+
"openlit",
|
371
|
+
SemanticConvetion.GEN_AI_APPLICATION_NAME:
|
372
|
+
application_name,
|
373
|
+
SemanticConvetion.GEN_AI_SYSTEM:
|
374
|
+
SemanticConvetion.GEN_AI_SYSTEM_COHERE,
|
375
|
+
SemanticConvetion.GEN_AI_ENVIRONMENT:
|
376
|
+
environment,
|
377
|
+
SemanticConvetion.GEN_AI_TYPE:
|
378
|
+
SemanticConvetion.GEN_AI_TYPE_CHAT,
|
379
|
+
SemanticConvetion.GEN_AI_REQUEST_MODEL:
|
380
|
+
kwargs.get("model", "command")
|
381
|
+
}
|
382
|
+
|
383
|
+
metrics["genai_requests"].add(1, attributes)
|
384
|
+
metrics["genai_total_tokens"].add(
|
385
|
+
prompt_tokens + completion_tokens, attributes
|
386
|
+
)
|
387
|
+
metrics["genai_completion_tokens"].add(completion_tokens, attributes)
|
388
|
+
metrics["genai_prompt_tokens"].add(prompt_tokens, attributes)
|
389
|
+
metrics["genai_cost"].record(cost, attributes)
|
390
|
+
|
391
|
+
except Exception as e:
|
392
|
+
handle_exception(span, e)
|
393
|
+
logger.error("Error in trace creation: %s", e)
|
394
|
+
|
395
|
+
return stream_generator()
|
396
|
+
|
397
|
+
return wrapper
|
@@ -0,0 +1,74 @@
|
|
1
|
+
# pylint: disable=useless-return, bad-staticmethod-argument, disable=duplicate-code
|
2
|
+
"""Initializer of Auto Instrumentation of LangChain Functions"""
|
3
|
+
from typing import Collection
|
4
|
+
import importlib.metadata
|
5
|
+
from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
|
6
|
+
from wrapt import wrap_function_wrapper
|
7
|
+
|
8
|
+
from openlit.instrumentation.langchain.langchain import general_wrap, hub
|
9
|
+
|
10
|
+
_instruments = ("langchain >= 0.1.1", "langchain-openai >= 0.1.1",
|
11
|
+
"langchain-core > 0.1.1", "langchain-community >= 0.0.31")
|
12
|
+
|
13
|
+
WRAPPED_METHODS = [
|
14
|
+
{
|
15
|
+
"package": "langchain_community.document_loaders.base",
|
16
|
+
"object": "BaseLoader.load",
|
17
|
+
"endpoint": "langchain.retrieve.load",
|
18
|
+
"wrapper": general_wrap,
|
19
|
+
},
|
20
|
+
{
|
21
|
+
"package": "langchain_community.document_loaders.base",
|
22
|
+
"object": "BaseLoader.aload",
|
23
|
+
"endpoint": "langchain.retrieve.load",
|
24
|
+
"wrapper": general_wrap,
|
25
|
+
},
|
26
|
+
{
|
27
|
+
"package": "langchain_text_splitters.base",
|
28
|
+
"object": "TextSplitter.split_documents",
|
29
|
+
"endpoint": "langchain.retrieve.split_documents",
|
30
|
+
"wrapper": general_wrap,
|
31
|
+
},
|
32
|
+
{
|
33
|
+
"package": "langchain_text_splitters.base",
|
34
|
+
"object": "TextSplitter.create_documents",
|
35
|
+
"endpoint": "langchain.retrieve.create_documents",
|
36
|
+
"wrapper": general_wrap,
|
37
|
+
},
|
38
|
+
{
|
39
|
+
"package": "langchain.hub",
|
40
|
+
"object": "pull",
|
41
|
+
"endpoint": "langchain.retrieve.prompt",
|
42
|
+
"wrapper": hub,
|
43
|
+
},
|
44
|
+
]
|
45
|
+
|
46
|
+
class LangChainInstrumentor(BaseInstrumentor):
|
47
|
+
"""An instrumentor for Cohere's client library."""
|
48
|
+
|
49
|
+
def instrumentation_dependencies(self) -> Collection[str]:
|
50
|
+
return _instruments
|
51
|
+
|
52
|
+
def _instrument(self, **kwargs):
|
53
|
+
application_name = kwargs.get("application_name")
|
54
|
+
environment = kwargs.get("environment")
|
55
|
+
tracer = kwargs.get("tracer")
|
56
|
+
pricing_info = kwargs.get("pricing_info")
|
57
|
+
trace_content = kwargs.get("trace_content")
|
58
|
+
version = importlib.metadata.version("langchain")
|
59
|
+
|
60
|
+
for wrapped_method in WRAPPED_METHODS:
|
61
|
+
wrap_package = wrapped_method.get("package")
|
62
|
+
wrap_object = wrapped_method.get("object")
|
63
|
+
gen_ai_endpoint = wrapped_method.get("endpoint")
|
64
|
+
wrapper = wrapped_method.get("wrapper")
|
65
|
+
wrap_function_wrapper(
|
66
|
+
wrap_package,
|
67
|
+
wrap_object,
|
68
|
+
wrapper(gen_ai_endpoint, version, environment, application_name,
|
69
|
+
tracer, pricing_info, trace_content),
|
70
|
+
)
|
71
|
+
|
72
|
+
@staticmethod
|
73
|
+
def _uninstrument(self, **kwargs):
|
74
|
+
pass
|
@@ -0,0 +1,161 @@
|
|
1
|
+
# pylint: disable=duplicate-code, broad-exception-caught, too-many-statements, unused-argument
|
2
|
+
"""
|
3
|
+
Module for monitoring Langchain applications.
|
4
|
+
"""
|
5
|
+
|
6
|
+
import logging
|
7
|
+
from opentelemetry.trace import SpanKind, Status, StatusCode
|
8
|
+
from opentelemetry.sdk.resources import TELEMETRY_SDK_NAME
|
9
|
+
from openlit.__helpers import handle_exception
|
10
|
+
from openlit.semcov import SemanticConvetion
|
11
|
+
|
12
|
+
# Initialize logger for logging potential issues and operations
|
13
|
+
logger = logging.getLogger(__name__)
|
14
|
+
|
15
|
+
def general_wrap(gen_ai_endpoint, version, environment, application_name,
|
16
|
+
tracer, pricing_info, trace_content):
|
17
|
+
"""
|
18
|
+
Creates a wrapper around a function call to trace and log its execution metrics.
|
19
|
+
|
20
|
+
This function wraps any given function to measure its execution time,
|
21
|
+
log its operation, and trace its execution using OpenTelemetry.
|
22
|
+
|
23
|
+
Parameters:
|
24
|
+
- gen_ai_endpoint (str): A descriptor or name for the endpoint being traced.
|
25
|
+
- version (str): The version of the Langchain application.
|
26
|
+
- environment (str): The deployment environment (e.g., 'production', 'development').
|
27
|
+
- application_name (str): Name of the Langchain application.
|
28
|
+
- tracer (opentelemetry.trace.Tracer): The tracer object used for OpenTelemetry tracing.
|
29
|
+
- pricing_info (dict): Information about the pricing for internal metrics (currently not used).
|
30
|
+
- trace_content (bool): Flag indicating whether to trace the content of the response.
|
31
|
+
|
32
|
+
Returns:
|
33
|
+
- function: A higher-order function that takes a function 'wrapped' and returns
|
34
|
+
a new function that wraps 'wrapped' with additional tracing and logging.
|
35
|
+
"""
|
36
|
+
|
37
|
+
def wrapper(wrapped, instance, args, kwargs):
|
38
|
+
"""
|
39
|
+
An inner wrapper function that executes the wrapped function, measures execution
|
40
|
+
time, and records trace data using OpenTelemetry.
|
41
|
+
|
42
|
+
Parameters:
|
43
|
+
- wrapped (Callable): The original function that this wrapper will execute.
|
44
|
+
- instance (object): The instance to which the wrapped function belongs. This
|
45
|
+
is used for instance methods. For static and classmethods,
|
46
|
+
this may be None.
|
47
|
+
- args (tuple): Positional arguments passed to the wrapped function.
|
48
|
+
- kwargs (dict): Keyword arguments passed to the wrapped function.
|
49
|
+
|
50
|
+
Returns:
|
51
|
+
- The result of the wrapped function call.
|
52
|
+
|
53
|
+
The wrapper initiates a span with the provided tracer, sets various attributes
|
54
|
+
on the span based on the function's execution and response, and ensures
|
55
|
+
errors are handled and logged appropriately.
|
56
|
+
"""
|
57
|
+
with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
|
58
|
+
response = wrapped(*args, **kwargs)
|
59
|
+
|
60
|
+
try:
|
61
|
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
|
62
|
+
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
|
63
|
+
"langchain")
|
64
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
|
65
|
+
gen_ai_endpoint)
|
66
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
|
67
|
+
environment)
|
68
|
+
span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
|
69
|
+
SemanticConvetion.GEN_AI_TYPE_FRAMEWORK)
|
70
|
+
span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
|
71
|
+
application_name)
|
72
|
+
span.set_attribute(SemanticConvetion.GEN_AI_RETRIEVAL_SOURCE,
|
73
|
+
response[0].metadata["source"])
|
74
|
+
span.set_status(Status(StatusCode.OK))
|
75
|
+
|
76
|
+
# Return original response
|
77
|
+
return response
|
78
|
+
|
79
|
+
except Exception as e:
|
80
|
+
handle_exception(span, e)
|
81
|
+
logger.error("Error in trace creation: %s", e)
|
82
|
+
|
83
|
+
# Return original response
|
84
|
+
return response
|
85
|
+
|
86
|
+
return wrapper
|
87
|
+
|
88
|
+
def hub(gen_ai_endpoint, version, environment, application_name, tracer,
|
89
|
+
pricing_info, trace_content):
|
90
|
+
"""
|
91
|
+
Creates a wrapper around Langchain hub operations for tracing and logging.
|
92
|
+
|
93
|
+
Similar to `general_wrap`, this function focuses on wrapping functions involved
|
94
|
+
in interacting with the Langchain hub, adding specific metadata relevant to
|
95
|
+
hub operations to the span attributes.
|
96
|
+
|
97
|
+
Parameters:
|
98
|
+
- gen_ai_endpoint (str): A descriptor or name for the Langchain hub endpoint.
|
99
|
+
- version (str): The version of the Langchain application.
|
100
|
+
- environment (str): The deployment environment, such as 'production' or 'development'.
|
101
|
+
- application_name (str): Name of the Langchain application.
|
102
|
+
- tracer (opentelemetry.trace.Tracer): The tracer for OpenTelemetry tracing.
|
103
|
+
- pricing_info (dict): Pricing information for the operation (not currently used).
|
104
|
+
- trace_content (bool): Indicates if the content of the response should be traced.
|
105
|
+
|
106
|
+
Returns:
|
107
|
+
- function: A new function that wraps the original hub operation call with added
|
108
|
+
logging, tracing, and metric calculation functionalities.
|
109
|
+
"""
|
110
|
+
|
111
|
+
def wrapper(wrapped, instance, args, kwargs):
|
112
|
+
"""
|
113
|
+
An inner wrapper specifically designed for Langchain hub operations,
|
114
|
+
providing tracing, logging, and execution metrics.
|
115
|
+
|
116
|
+
Parameters:
|
117
|
+
- wrapped (Callable): The original hub operation function to be executed.
|
118
|
+
- instance (object): The instance of the class where the hub operation
|
119
|
+
method is defined. May be None for static or class methods.
|
120
|
+
- args (tuple): Positional arguments to pass to the hub operation function.
|
121
|
+
- kwargs (dict): Keyword arguments to pass to the hub operation function.
|
122
|
+
|
123
|
+
Returns:
|
124
|
+
- The result of executing the hub operation function.
|
125
|
+
|
126
|
+
This wrapper captures additional metadata relevant to Langchain hub operations,
|
127
|
+
creating spans with specific attributes and metrics that reflect the nature of
|
128
|
+
each hub call.
|
129
|
+
"""
|
130
|
+
|
131
|
+
with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
|
132
|
+
response = wrapped(*args, **kwargs)
|
133
|
+
|
134
|
+
try:
|
135
|
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
|
136
|
+
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
|
137
|
+
SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN)
|
138
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
|
139
|
+
gen_ai_endpoint)
|
140
|
+
span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
|
141
|
+
environment)
|
142
|
+
span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
|
143
|
+
SemanticConvetion.GEN_AI_TYPE_FRAMEWORK)
|
144
|
+
span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
|
145
|
+
application_name)
|
146
|
+
span.set_attribute(SemanticConvetion.GEN_AI_HUB_OWNER,
|
147
|
+
response.metadata["lc_hub_owner"])
|
148
|
+
span.set_attribute(SemanticConvetion.GEN_AI_HUB_REPO,
|
149
|
+
response.metadata["lc_hub_repo"])
|
150
|
+
span.set_status(Status(StatusCode.OK))
|
151
|
+
|
152
|
+
return response
|
153
|
+
|
154
|
+
except Exception as e:
|
155
|
+
handle_exception(span, e)
|
156
|
+
logger.error("Error in trace creation: %s", e)
|
157
|
+
|
158
|
+
# Return original response
|
159
|
+
return response
|
160
|
+
|
161
|
+
return wrapper
|