langwatch 0.1.1 → 0.1.4
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.
- package/copy-types.sh +17 -0
- package/dist/{chunk-OVS4NSDE.mjs → chunk-2I4YLOQY.mjs} +184 -115
- package/dist/chunk-2I4YLOQY.mjs.map +1 -0
- package/dist/index.d.mts +344 -4
- package/dist/index.d.ts +344 -4
- package/dist/index.js +1429 -1121
- package/dist/index.js.map +1 -1
- package/dist/index.mjs +264 -26
- package/dist/index.mjs.map +1 -1
- package/dist/{utils-K-jSEpnZ.d.mts → utils-CFtM8VVg.d.mts} +107 -31
- package/dist/{utils-K-jSEpnZ.d.ts → utils-CFtM8VVg.d.ts} +107 -31
- package/dist/utils.d.mts +1 -2
- package/dist/utils.d.ts +1 -2
- package/dist/utils.js +181 -114
- package/dist/utils.js.map +1 -1
- package/dist/utils.mjs +1 -1
- package/example/app/guardrails/page.tsx +26 -0
- package/example/components/header.tsx +4 -0
- package/example/lib/chat/guardrails.tsx +181 -0
- package/example/lib/chat/langchain-rag.tsx +1 -1
- package/example/lib/chat/langchain.tsx +1 -1
- package/example/lib/chat/vercel-ai.tsx +1 -1
- package/example/package-lock.json +4 -3
- package/package.json +3 -4
- package/src/evaluations.ts +219 -0
- package/src/index.test.ts +40 -2
- package/src/index.ts +182 -8
- package/src/typeUtils.ts +28 -3
- package/src/types.ts +6 -2
- package/src/utils.ts +4 -8
- package/ts-to-zod.config.js +4 -0
- package/dist/chunk-OVS4NSDE.mjs.map +0 -1
package/dist/index.d.ts
CHANGED
|
@@ -1,6 +1,4 @@
|
|
|
1
1
|
import EventEmitter from 'events';
|
|
2
|
-
import { M as Metadata, C as CollectorRESTParams, S as Span, P as PendingBaseSpan, a as PendingLLMSpan, b as PendingRAGSpan, c as SpanTypes } from './utils-K-jSEpnZ.js';
|
|
3
|
-
export { B as BaseSpan, d as ChatMessage, e as ChatRichContent, L as LLMSpan, R as RAGSpan, f as SpanInputOutput, i as autoconvertTypedValues, h as captureError, g as convertFromVercelAIMessages } from './utils-K-jSEpnZ.js';
|
|
4
2
|
import { AgentAction, AgentFinish } from '@langchain/core/agents';
|
|
5
3
|
import { BaseCallbackHandler } from '@langchain/core/callbacks/base';
|
|
6
4
|
import { DocumentInterface } from '@langchain/core/documents';
|
|
@@ -8,8 +6,327 @@ import { Serialized } from '@langchain/core/load/serializable';
|
|
|
8
6
|
import { BaseMessage } from '@langchain/core/messages';
|
|
9
7
|
import { LLMResult } from '@langchain/core/outputs';
|
|
10
8
|
import { ChainValues } from '@langchain/core/utils/types';
|
|
9
|
+
import { R as RAGChunk, M as Metadata, C as CollectorRESTParams, S as Span, a as RESTEvaluation, P as PendingBaseSpan, b as PendingLLMSpan, c as PendingRAGSpan, d as SpanTypes } from './utils-CFtM8VVg.js';
|
|
10
|
+
export { B as BaseSpan, e as ChatMessage, f as ChatRichContent, L as LLMSpan, g as RAGSpan, h as SpanInputOutput, i as autoconvertTypedValues, j as captureError, k as convertFromVercelAIMessages } from './utils-CFtM8VVg.js';
|
|
11
11
|
import 'ai';
|
|
12
|
-
|
|
12
|
+
|
|
13
|
+
type EvaluatorTypes = keyof Evaluators;
|
|
14
|
+
type Evaluators = {
|
|
15
|
+
"lingua/language_detection": {
|
|
16
|
+
settings: {
|
|
17
|
+
check_for: "input_matches_output" | "output_matches_language";
|
|
18
|
+
expected_language?: "AF" | "AR" | "AZ" | "BE" | "BG" | "BN" | "BS" | "CA" | "CS" | "CY" | "DA" | "DE" | "EL" | "EN" | "EO" | "ES" | "ET" | "EU" | "FA" | "FI" | "FR" | "GA" | "GU" | "HE" | "HI" | "HR" | "HU" | "HY" | "ID" | "IS" | "IT" | "JA" | "KA" | "KK" | "KO" | "LA" | "LG" | "LT" | "LV" | "MI" | "MK" | "MN" | "MR" | "MS" | "NB" | "NL" | "NN" | "PA" | "PL" | "PT" | "RO" | "RU" | "SK" | "SL" | "SN" | "SO" | "SQ" | "SR" | "ST" | "SV" | "SW" | "TA" | "TE" | "TH" | "TL" | "TN" | "TR" | "TS" | "UK" | "UR" | "VI" | "XH" | "YO" | "ZH" | "ZU";
|
|
19
|
+
min_words: number;
|
|
20
|
+
threshold: number;
|
|
21
|
+
};
|
|
22
|
+
};
|
|
23
|
+
"aws/comprehend_pii_detection": {
|
|
24
|
+
settings: {
|
|
25
|
+
entity_types: {
|
|
26
|
+
BANK_ACCOUNT_NUMBER: boolean;
|
|
27
|
+
BANK_ROUTING: boolean;
|
|
28
|
+
CREDIT_DEBIT_NUMBER: boolean;
|
|
29
|
+
CREDIT_DEBIT_CVV: boolean;
|
|
30
|
+
CREDIT_DEBIT_EXPIRY: boolean;
|
|
31
|
+
PIN: boolean;
|
|
32
|
+
EMAIL: boolean;
|
|
33
|
+
ADDRESS: boolean;
|
|
34
|
+
NAME: boolean;
|
|
35
|
+
PHONE: boolean;
|
|
36
|
+
SSN: boolean;
|
|
37
|
+
DATE_TIME: boolean;
|
|
38
|
+
PASSPORT_NUMBER: boolean;
|
|
39
|
+
DRIVER_ID: boolean;
|
|
40
|
+
URL: boolean;
|
|
41
|
+
AGE: boolean;
|
|
42
|
+
USERNAME: boolean;
|
|
43
|
+
PASSWORD: boolean;
|
|
44
|
+
AWS_ACCESS_KEY: boolean;
|
|
45
|
+
AWS_SECRET_KEY: boolean;
|
|
46
|
+
IP_ADDRESS: boolean;
|
|
47
|
+
MAC_ADDRESS: boolean;
|
|
48
|
+
LICENSE_PLATE: boolean;
|
|
49
|
+
VEHICLE_IDENTIFICATION_NUMBER: boolean;
|
|
50
|
+
UK_NATIONAL_INSURANCE_NUMBER: boolean;
|
|
51
|
+
CA_SOCIAL_INSURANCE_NUMBER: boolean;
|
|
52
|
+
US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER: boolean;
|
|
53
|
+
UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER: boolean;
|
|
54
|
+
IN_PERMANENT_ACCOUNT_NUMBER: boolean;
|
|
55
|
+
IN_NREGA: boolean;
|
|
56
|
+
INTERNATIONAL_BANK_ACCOUNT_NUMBER: boolean;
|
|
57
|
+
SWIFT_CODE: boolean;
|
|
58
|
+
UK_NATIONAL_HEALTH_SERVICE_NUMBER: boolean;
|
|
59
|
+
CA_HEALTH_NUMBER: boolean;
|
|
60
|
+
IN_AADHAAR: boolean;
|
|
61
|
+
IN_VOTER_NUMBER: boolean;
|
|
62
|
+
};
|
|
63
|
+
language_code: "en" | "es" | "fr" | "de" | "it" | "pt" | "ar" | "hi" | "ja" | "ko" | "zh" | "zh-TW";
|
|
64
|
+
min_confidence: number;
|
|
65
|
+
aws_region: "us-east-1" | "us-east-2" | "us-west-1" | "us-west-2" | "ap-east-1" | "ap-south-1" | "ap-northeast-3" | "ap-northeast-2" | "ap-southeast-1" | "ap-southeast-2" | "ap-northeast-1" | "ca-central-1" | "eu-central-1" | "eu-west-1" | "eu-west-2" | "eu-south-1" | "eu-west-3" | "eu-north-1" | "me-south-1" | "sa-east-1";
|
|
66
|
+
};
|
|
67
|
+
};
|
|
68
|
+
"huggingface/llama_guard": {
|
|
69
|
+
settings: {
|
|
70
|
+
policy: string;
|
|
71
|
+
evaluate: "input" | "output" | "both";
|
|
72
|
+
model: "cloudflare/thebloke/llamaguard-7b-awq";
|
|
73
|
+
};
|
|
74
|
+
};
|
|
75
|
+
"google_cloud/dlp_pii_detection": {
|
|
76
|
+
settings: {
|
|
77
|
+
info_types: {
|
|
78
|
+
phone_number: boolean;
|
|
79
|
+
email_address: boolean;
|
|
80
|
+
credit_card_number: boolean;
|
|
81
|
+
iban_code: boolean;
|
|
82
|
+
ip_address: boolean;
|
|
83
|
+
passport: boolean;
|
|
84
|
+
vat_number: boolean;
|
|
85
|
+
medical_record_number: boolean;
|
|
86
|
+
};
|
|
87
|
+
min_likelihood: "VERY_UNLIKELY" | "UNLIKELY" | "POSSIBLE" | "LIKELY" | "VERY_LIKELY";
|
|
88
|
+
};
|
|
89
|
+
};
|
|
90
|
+
"presidio/pii_detection": {
|
|
91
|
+
settings: {
|
|
92
|
+
entities: {
|
|
93
|
+
credit_card: boolean;
|
|
94
|
+
crypto: boolean;
|
|
95
|
+
email_address: boolean;
|
|
96
|
+
iban_code: boolean;
|
|
97
|
+
ip_address: boolean;
|
|
98
|
+
location: boolean;
|
|
99
|
+
person: boolean;
|
|
100
|
+
phone_number: boolean;
|
|
101
|
+
medical_license: boolean;
|
|
102
|
+
us_bank_number: boolean;
|
|
103
|
+
us_driver_license: boolean;
|
|
104
|
+
us_itin: boolean;
|
|
105
|
+
us_passport: boolean;
|
|
106
|
+
us_ssn: boolean;
|
|
107
|
+
uk_nhs: boolean;
|
|
108
|
+
sg_nric_fin: boolean;
|
|
109
|
+
sg_uen: boolean;
|
|
110
|
+
au_abn: boolean;
|
|
111
|
+
au_acn: boolean;
|
|
112
|
+
au_tfn: boolean;
|
|
113
|
+
au_medicare: boolean;
|
|
114
|
+
in_pan: boolean;
|
|
115
|
+
in_aadhaar: boolean;
|
|
116
|
+
in_vehicle_registration: boolean;
|
|
117
|
+
in_voter: boolean;
|
|
118
|
+
in_passport: boolean;
|
|
119
|
+
fi_personal_identity_code: boolean;
|
|
120
|
+
};
|
|
121
|
+
min_threshold: number;
|
|
122
|
+
};
|
|
123
|
+
};
|
|
124
|
+
"ragas/answer_correctness": {
|
|
125
|
+
settings: {
|
|
126
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
127
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
128
|
+
max_tokens: number;
|
|
129
|
+
};
|
|
130
|
+
};
|
|
131
|
+
"ragas/answer_relevancy": {
|
|
132
|
+
settings: {
|
|
133
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
134
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
135
|
+
max_tokens: number;
|
|
136
|
+
};
|
|
137
|
+
};
|
|
138
|
+
"ragas/context_precision": {
|
|
139
|
+
settings: {
|
|
140
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
141
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
142
|
+
max_tokens: number;
|
|
143
|
+
};
|
|
144
|
+
};
|
|
145
|
+
"ragas/context_recall": {
|
|
146
|
+
settings: {
|
|
147
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
148
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
149
|
+
max_tokens: number;
|
|
150
|
+
};
|
|
151
|
+
};
|
|
152
|
+
"ragas/context_relevancy": {
|
|
153
|
+
settings: {
|
|
154
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
155
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
156
|
+
max_tokens: number;
|
|
157
|
+
};
|
|
158
|
+
};
|
|
159
|
+
"ragas/context_utilization": {
|
|
160
|
+
settings: {
|
|
161
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
162
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
163
|
+
max_tokens: number;
|
|
164
|
+
};
|
|
165
|
+
};
|
|
166
|
+
"ragas/faithfulness": {
|
|
167
|
+
settings: {
|
|
168
|
+
model: "openai/gpt-3.5-turbo-16k" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "azure/gpt-35-turbo-16k" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "anthropic/claude-3-5-sonnet-20240620";
|
|
169
|
+
embeddings_model: "openai/text-embedding-ada-002" | "azure/text-embedding-ada-002";
|
|
170
|
+
max_tokens: number;
|
|
171
|
+
};
|
|
172
|
+
};
|
|
173
|
+
"langevals/basic": {
|
|
174
|
+
settings: {
|
|
175
|
+
rules: {
|
|
176
|
+
field: "input" | "output";
|
|
177
|
+
rule: "contains" | "not_contains" | "matches_regex" | "not_matches_regex";
|
|
178
|
+
value: string;
|
|
179
|
+
}[];
|
|
180
|
+
};
|
|
181
|
+
};
|
|
182
|
+
"langevals/competitor_blocklist": {
|
|
183
|
+
settings: {
|
|
184
|
+
competitors: string[];
|
|
185
|
+
};
|
|
186
|
+
};
|
|
187
|
+
"langevals/competitor_llm": {
|
|
188
|
+
settings: {
|
|
189
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
190
|
+
max_tokens: number;
|
|
191
|
+
name: string;
|
|
192
|
+
description: string;
|
|
193
|
+
};
|
|
194
|
+
};
|
|
195
|
+
"langevals/competitor_llm_function_call": {
|
|
196
|
+
settings: {
|
|
197
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
198
|
+
max_tokens: number;
|
|
199
|
+
name: string;
|
|
200
|
+
description: string;
|
|
201
|
+
competitors: string[];
|
|
202
|
+
};
|
|
203
|
+
};
|
|
204
|
+
"langevals/llm_boolean": {
|
|
205
|
+
settings: {
|
|
206
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
207
|
+
max_tokens: number;
|
|
208
|
+
prompt: string;
|
|
209
|
+
};
|
|
210
|
+
};
|
|
211
|
+
"langevals/llm_score": {
|
|
212
|
+
settings: {
|
|
213
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
214
|
+
max_tokens: number;
|
|
215
|
+
prompt: string;
|
|
216
|
+
};
|
|
217
|
+
};
|
|
218
|
+
"langevals/off_topic": {
|
|
219
|
+
settings: {
|
|
220
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
221
|
+
max_tokens: number;
|
|
222
|
+
allowed_topics: {
|
|
223
|
+
topic: string;
|
|
224
|
+
description: string;
|
|
225
|
+
}[];
|
|
226
|
+
};
|
|
227
|
+
};
|
|
228
|
+
"langevals/product_sentiment_polarity": {
|
|
229
|
+
settings: Record<string, never>;
|
|
230
|
+
};
|
|
231
|
+
"langevals/query_resolution": {
|
|
232
|
+
settings: {
|
|
233
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
234
|
+
max_tokens: number;
|
|
235
|
+
};
|
|
236
|
+
};
|
|
237
|
+
"langevals/similarity": {
|
|
238
|
+
settings: {
|
|
239
|
+
field: "input" | "output";
|
|
240
|
+
rule: "is_not_similar_to" | "is_similar_to";
|
|
241
|
+
value: string;
|
|
242
|
+
threshold: number;
|
|
243
|
+
embeddings_model: "openai/text-embedding-3-small" | "azure/text-embedding-ada-002";
|
|
244
|
+
};
|
|
245
|
+
};
|
|
246
|
+
"azure/content_safety": {
|
|
247
|
+
settings: {
|
|
248
|
+
severity_threshold: 1 | 2 | 3 | 4 | 5 | 6 | 7;
|
|
249
|
+
categories: {
|
|
250
|
+
Hate: boolean;
|
|
251
|
+
SelfHarm: boolean;
|
|
252
|
+
Sexual: boolean;
|
|
253
|
+
Violence: boolean;
|
|
254
|
+
};
|
|
255
|
+
output_type: "FourSeverityLevels" | "EightSeverityLevels";
|
|
256
|
+
};
|
|
257
|
+
};
|
|
258
|
+
"azure/jailbreak": {
|
|
259
|
+
settings: Record<string, never>;
|
|
260
|
+
};
|
|
261
|
+
"azure/prompt_injection": {
|
|
262
|
+
settings: Record<string, never>;
|
|
263
|
+
};
|
|
264
|
+
"openai/moderation": {
|
|
265
|
+
settings: {
|
|
266
|
+
model: "text-moderation-stable" | "text-moderation-latest";
|
|
267
|
+
categories: {
|
|
268
|
+
harassment: boolean;
|
|
269
|
+
harassment_threatening: boolean;
|
|
270
|
+
hate: boolean;
|
|
271
|
+
hate_threatening: boolean;
|
|
272
|
+
self_harm: boolean;
|
|
273
|
+
self_harm_instructions: boolean;
|
|
274
|
+
self_harm_intent: boolean;
|
|
275
|
+
sexual: boolean;
|
|
276
|
+
sexual_minors: boolean;
|
|
277
|
+
violence: boolean;
|
|
278
|
+
violence_graphic: boolean;
|
|
279
|
+
};
|
|
280
|
+
};
|
|
281
|
+
};
|
|
282
|
+
"example/word_count": {
|
|
283
|
+
settings: Record<string, never>;
|
|
284
|
+
};
|
|
285
|
+
"haystack/faithfulness": {
|
|
286
|
+
settings: {
|
|
287
|
+
model: "openai/gpt-3.5-turbo" | "openai/gpt-3.5-turbo-0125" | "openai/gpt-3.5-turbo-1106" | "openai/gpt-4-turbo" | "openai/gpt-4-0125-preview" | "openai/gpt-4o" | "openai/gpt-4o-mini" | "openai/gpt-4-1106-preview" | "azure/gpt-35-turbo-1106" | "azure/gpt-4o" | "azure/gpt-4o-mini" | "azure/gpt-4-turbo-2024-04-09" | "azure/gpt-4-1106-preview" | "groq/llama3-70b-8192" | "anthropic/claude-3-haiku-20240307" | "anthropic/claude-3-5-sonnet-20240620" | "anthropic/claude-3-opus-20240229";
|
|
288
|
+
max_tokens: number;
|
|
289
|
+
};
|
|
290
|
+
};
|
|
291
|
+
};
|
|
292
|
+
|
|
293
|
+
type Conversation = {
|
|
294
|
+
input?: string;
|
|
295
|
+
output?: string;
|
|
296
|
+
}[];
|
|
297
|
+
|
|
298
|
+
type Money = {
|
|
299
|
+
currency: string;
|
|
300
|
+
amount: number;
|
|
301
|
+
};
|
|
302
|
+
type EvaluationResultModel = {
|
|
303
|
+
status: "processed" | "skipped" | "error";
|
|
304
|
+
passed?: boolean;
|
|
305
|
+
score?: number;
|
|
306
|
+
details?: string;
|
|
307
|
+
label?: string;
|
|
308
|
+
cost?: Money;
|
|
309
|
+
};
|
|
310
|
+
type CommonEvaluationParams = {
|
|
311
|
+
name?: string;
|
|
312
|
+
input?: string;
|
|
313
|
+
output?: string;
|
|
314
|
+
expectedOutput?: string;
|
|
315
|
+
contexts?: RAGChunk[] | string[];
|
|
316
|
+
conversation?: Conversation;
|
|
317
|
+
asGuardrail?: boolean;
|
|
318
|
+
trace?: LangWatchTrace;
|
|
319
|
+
span?: LangWatchSpan;
|
|
320
|
+
};
|
|
321
|
+
type SavedEvaluationParams = {
|
|
322
|
+
slug: string;
|
|
323
|
+
settings?: Record<string, unknown>;
|
|
324
|
+
} & CommonEvaluationParams;
|
|
325
|
+
type LangEvalsEvaluationParams<T extends EvaluatorTypes> = {
|
|
326
|
+
evaluator: T;
|
|
327
|
+
settings?: Evaluators[T]["settings"];
|
|
328
|
+
} & CommonEvaluationParams;
|
|
329
|
+
type EvaluationParams = SavedEvaluationParams | LangEvalsEvaluationParams<EvaluatorTypes>;
|
|
13
330
|
|
|
14
331
|
declare class LangWatchCallbackHandler extends BaseCallbackHandler {
|
|
15
332
|
name: string;
|
|
@@ -58,13 +375,29 @@ declare class LangWatch extends EventEmitter {
|
|
|
58
375
|
sendTrace(params: CollectorRESTParams): Promise<void>;
|
|
59
376
|
_sendTrace(params: CollectorRESTParams): Promise<void>;
|
|
60
377
|
}
|
|
378
|
+
type AddEvaluationParams = {
|
|
379
|
+
evaluationId?: string;
|
|
380
|
+
span?: LangWatchSpan;
|
|
381
|
+
name: string;
|
|
382
|
+
type?: string;
|
|
383
|
+
isGuardrail?: boolean;
|
|
384
|
+
status?: "processed" | "skipped" | "error";
|
|
385
|
+
passed?: boolean;
|
|
386
|
+
score?: number;
|
|
387
|
+
label?: string;
|
|
388
|
+
details?: string;
|
|
389
|
+
error?: Error;
|
|
390
|
+
timestamps?: RESTEvaluation["timestamps"];
|
|
391
|
+
};
|
|
61
392
|
declare class LangWatchTrace {
|
|
62
393
|
client: LangWatch;
|
|
63
394
|
traceId: string;
|
|
64
395
|
metadata?: Metadata;
|
|
65
396
|
finishedSpans: Record<string, Span>;
|
|
66
|
-
timeoutRef?: NodeJS.Timeout;
|
|
67
397
|
langchainCallback?: LangWatchCallbackHandler;
|
|
398
|
+
evaluations: RESTEvaluation[];
|
|
399
|
+
private currentSpan?;
|
|
400
|
+
private timeoutRef?;
|
|
68
401
|
constructor({ client, traceId, metadata, }: {
|
|
69
402
|
client: LangWatch;
|
|
70
403
|
traceId: string;
|
|
@@ -73,9 +406,14 @@ declare class LangWatchTrace {
|
|
|
73
406
|
update({ metadata }: {
|
|
74
407
|
metadata: Metadata;
|
|
75
408
|
}): void;
|
|
409
|
+
setCurrentSpan(span: LangWatchSpan): void;
|
|
410
|
+
getCurrentSpan(): LangWatchSpan | undefined;
|
|
411
|
+
resetCurrentSpan(): void;
|
|
76
412
|
startSpan(params: Omit<Partial<PendingBaseSpan>, "parentId">): LangWatchSpan;
|
|
77
413
|
startLLMSpan(params: Omit<Partial<PendingLLMSpan>, "parentId">): LangWatchLLMSpan;
|
|
78
414
|
startRAGSpan(params: Omit<Partial<PendingRAGSpan>, "parentId">): LangWatchRAGSpan;
|
|
415
|
+
addEvaluation: ({ evaluationId, span, name, type, isGuardrail, status, passed, score, label, details, error, timestamps, }: AddEvaluationParams) => void;
|
|
416
|
+
evaluate(params: EvaluationParams): Promise<EvaluationResultModel>;
|
|
79
417
|
getLangChainCallback(): LangWatchCallbackHandler;
|
|
80
418
|
onEnd(span: Span): void;
|
|
81
419
|
delayedSendSpans(): void;
|
|
@@ -99,6 +437,8 @@ declare class LangWatchSpan implements PendingBaseSpan {
|
|
|
99
437
|
startSpan(params: Omit<Partial<PendingBaseSpan>, "parentId">): LangWatchSpan;
|
|
100
438
|
startLLMSpan(params: Omit<Partial<PendingLLMSpan>, "parentId">): LangWatchLLMSpan;
|
|
101
439
|
startRAGSpan(params: Omit<Partial<PendingRAGSpan>, "parentId">): LangWatchRAGSpan;
|
|
440
|
+
addEvaluation(params: AddEvaluationParams): void;
|
|
441
|
+
evaluate(params: EvaluationParams): Promise<EvaluationResultModel>;
|
|
102
442
|
end(params?: Partial<Omit<PendingBaseSpan, "spanId" | "parentId">>): void;
|
|
103
443
|
}
|
|
104
444
|
declare class LangWatchLLMSpan extends LangWatchSpan implements PendingLLMSpan {
|