Skip to content

feat: set up gen ai inference attributes for foundational text models #111

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Nov 1, 2024
Merged

feat: set up gen ai inference attributes for foundational text models #111

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
3aa360d
feat: set up gen ai inference attributes for foundational text models
yiyuan-he Oct 28, 2024
2061da2
Merge branch 'aws-observability:main' into gen-ai-support
yiyuan-he Oct 29, 2024
6cd9bcc
chore: apply lint
yiyuan-he Oct 29, 2024
842f9c9
feat: add unit tests for gen ai inference attributes
yiyuan-he Oct 29, 2024
c730602
Merge branch 'main' into gen-ai-support
yiyuan-he Oct 30, 2024
3a57302
fix: add strict undefined checks
yiyuan-he Oct 31, 2024
fae5c83
doc: add inline comments for token count approximation
yiyuan-he Oct 31, 2024
650eb93
chore: apply lint
yiyuan-he Oct 31, 2024
ad29de3
Merge branch 'main' into gen-ai-support
yiyuan-he Oct 31, 2024
63d6800
Merge branch 'main' into gen-ai-support
yiyuan-he Nov 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions aws-distro-opentelemetry-node-autoinstrumentation/src/aws-span-processing-util.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,12 @@ export class AwsSpanProcessingUtil {
// TODO: Use Semantic Conventions once upgraded
static GEN_AI_REQUEST_MODEL: string = 'gen_ai.request.model';
static GEN_AI_SYSTEM: string = 'gen_ai.system';
static GEN_AI_REQUEST_MAX_TOKENS: string = 'gen_ai.request.max_tokens';
static GEN_AI_REQUEST_TEMPERATURE: string = 'gen_ai.request.temperature';
static GEN_AI_REQUEST_TOP_P: string = 'gen_ai.request.top_p';
static GEN_AI_RESPONSE_FINISH_REASONS: string = 'gen_ai.response.finish_reasons';
static GEN_AI_USAGE_INPUT_TOKENS: string = 'gen_ai.usage.input_tokens';
static GEN_AI_USAGE_OUTPUT_TOKENS: string = 'gen_ai.usage.output_tokens';

static getDialectKeywords(): string[] {
return SQL_DIALECT_KEYWORDS_JSON.keywords;
Expand Down
145 changes: 145 additions & 0 deletions aws-distro-opentelemetry-node-autoinstrumentation/src/patches/aws/services/bedrock.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -211,11 +211,156 @@ export class BedrockRuntimeServiceExtension implements ServiceExtension {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MODEL] = modelId;
}

if (request.commandInput?.body) {
const requestBody = JSON.parse(request.commandInput.body);
if (modelId.includes('amazon.titan')) {
if (requestBody.textGenerationConfig) {
const config = requestBody.textGenerationConfig;
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = config.temperature;
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = config.topP;
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = config.maxTokenCount;
}
} else if (modelId.includes('anthropic.claude')) {
if (requestBody.max_tokens) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = requestBody.max_tokens;
}
if (requestBody.temperature) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = requestBody.temperature;
}
if (requestBody.top_p) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = requestBody.top_p;
}
} else if (modelId.includes('meta.llama')) {
if (requestBody.max_gen_len) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = requestBody.max_gen_len;
}
if (requestBody.temperature) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = requestBody.temperature;
}
if (requestBody.top_p) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = requestBody.top_p;
}
} else if (modelId.includes('cohere.command')) {
if (requestBody.max_tokens) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = requestBody.max_tokens;
}
if (requestBody.temperature) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = requestBody.temperature;
}
if (requestBody.p) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = requestBody.p;
}
} else if (modelId.includes('ai21.jamba')) {
if (requestBody.max_tokens) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = requestBody.max_tokens;
}
if (requestBody.temperature) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = requestBody.temperature;
}
if (requestBody.top_p) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = requestBody.top_p;
}
} else if (modelId.includes('mistral.mistral')) {
if (requestBody.prompt) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS] = Math.ceil(requestBody.prompt.length / 6);
}
if (requestBody.max_tokens) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_MAX_TOKENS] = requestBody.max_tokens;
}
if (requestBody.temperature) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TEMPERATURE] = requestBody.temperature;
}
if (requestBody.top_p) {
spanAttributes[AwsSpanProcessingUtil.GEN_AI_REQUEST_TOP_P] = requestBody.top_p;
}
}
}

return {
isIncoming,
spanAttributes,
spanKind,
spanName,
};
}

responseHook(response: NormalizedResponse, span: Span, tracer: Tracer, config: AwsSdkInstrumentationConfig): void {
const currentModelId = response.request.commandInput?.modelId;
if (response.data?.body) {
const decodedResponseBody = new TextDecoder().decode(response.data.body);
const responseBody = JSON.parse(decodedResponseBody);
if (currentModelId.includes('amazon.titan')) {
if (responseBody.inputTextTokenCount) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS, responseBody.inputTextTokenCount);
}
if (responseBody.results?.[0]?.tokenCount) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS, responseBody.results[0].tokenCount);
}
if (responseBody.results?.[0]?.completionReason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [
responseBody.results[0].completionReason,
]);
}
} else if (currentModelId.includes('anthropic.claude')) {
if (responseBody.usage?.input_tokens) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS, responseBody.usage.input_tokens);
}
if (responseBody.usage?.output_tokens) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS, responseBody.usage.output_tokens);
}
if (responseBody.stop_reason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [responseBody.stop_reason]);
}
} else if (currentModelId.includes('meta.llama')) {
if (responseBody.prompt_token_count) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS, responseBody.prompt_token_count);
}
if (responseBody.generation_token_count) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS, responseBody.generation_token_count);
}
if (responseBody.stop_reason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [responseBody.stop_reason]);
}
} else if (currentModelId.includes('cohere.command')) {
if (responseBody.prompt) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS, Math.ceil(responseBody.prompt.length / 6));
}
if (responseBody.generations?.[0]?.text) {
span.setAttribute(
AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS,
Math.ceil(responseBody.generations[0].text.length / 6)
);
}
if (responseBody.generations?.[0]?.finish_reason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [
responseBody.generations[0].finish_reason,
]);
}
} else if (currentModelId.includes('ai21.jamba')) {
if (responseBody.usage?.prompt_tokens) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_INPUT_TOKENS, responseBody.usage.prompt_tokens);
}
if (responseBody.usage?.completion_tokens) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS, responseBody.usage.completion_tokens);
}
if (responseBody.choices?.[0]?.finish_reason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [
responseBody.choices[0].finish_reason,
]);
}
} else if (currentModelId.includes('mistral.mistral')) {
if (responseBody.outputs?.[0]?.text) {
span.setAttribute(
AwsSpanProcessingUtil.GEN_AI_USAGE_OUTPUT_TOKENS,
Math.ceil(responseBody.outputs[0].text.length / 6)
);
}
if (responseBody.outputs?.[0]?.stop_reason) {
span.setAttribute(AwsSpanProcessingUtil.GEN_AI_RESPONSE_FINISH_REASONS, [
responseBody.outputs[0].stop_reason,
]);
}
}
}
}
}
Loading
Loading