feat: track token usage and estimated cost per filter run in filter_runs.json
This commit is contained in:
@@ -146,6 +146,14 @@ export async function checkBatch(batchId, apiKey) {
|
||||
/**
|
||||
* Download and parse batch results. Returns array of { jobId, score, reason, error }
|
||||
*/
|
||||
// Sonnet batch API pricing (per million tokens)
|
||||
const PRICING = {
|
||||
input: 1.50,
|
||||
output: 7.50,
|
||||
cache_write: 1.875,
|
||||
cache_read: 0.15,
|
||||
};
|
||||
|
||||
export async function downloadResults(batchId, apiKey, idMap = {}) {
|
||||
const res = await fetch(`${BATCH_API}/${batchId}/results`, {
|
||||
headers: apiHeaders(apiKey),
|
||||
@@ -156,13 +164,22 @@ export async function downloadResults(batchId, apiKey, idMap = {}) {
|
||||
const text = await res.text();
|
||||
const lines = text.trim().split('\n').filter(Boolean);
|
||||
const results = [];
|
||||
const usage = { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 };
|
||||
|
||||
for (const line of lines) {
|
||||
try {
|
||||
const entry = JSON.parse(line);
|
||||
// Resolve truncated custom_id back to original job ID
|
||||
const jobId = idMap[entry.custom_id] || entry.custom_id;
|
||||
|
||||
// Accumulate token usage
|
||||
const u = entry.result?.message?.usage;
|
||||
if (u) {
|
||||
usage.input_tokens += u.input_tokens || 0;
|
||||
usage.output_tokens += u.output_tokens || 0;
|
||||
usage.cache_creation_input_tokens += u.cache_creation_input_tokens || 0;
|
||||
usage.cache_read_input_tokens += u.cache_read_input_tokens || 0;
|
||||
}
|
||||
|
||||
if (entry.result?.type === 'succeeded') {
|
||||
const content = entry.result.message?.content?.[0]?.text || '';
|
||||
try {
|
||||
@@ -180,5 +197,13 @@ export async function downloadResults(batchId, apiKey, idMap = {}) {
|
||||
}
|
||||
}
|
||||
|
||||
return results;
|
||||
// Calculate estimated cost
|
||||
const cost = (
|
||||
(usage.input_tokens * PRICING.input) +
|
||||
(usage.output_tokens * PRICING.output) +
|
||||
(usage.cache_creation_input_tokens * PRICING.cache_write) +
|
||||
(usage.cache_read_input_tokens * PRICING.cache_read)
|
||||
) / 1_000_000;
|
||||
|
||||
return { results, usage, cost: Math.round(cost * 100) / 100 };
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user