245 lines
10 KiB
JSON
245 lines
10 KiB
JSON
{
|
|
"info": {
|
|
"name": "Bifrost V1 - Async Inference",
|
|
"description": "Async inference submit/poll tests. Requires LogsStore and governance plugin.",
|
|
"schema": "https://schema.getpostman.com/json/collection/v2.1.0/collection.json"
|
|
},
|
|
"variable": [
|
|
{"key": "base_url", "value": "http://localhost:8080", "type": "string"},
|
|
{"key": "provider", "value": "openai", "type": "string"},
|
|
{"key": "chat_model", "value": "gpt-4o", "type": "string"},
|
|
{"key": "embedding_model", "value": "text-embedding-3-small", "type": "string"},
|
|
{"key": "job_id", "value": "", "type": "string"},
|
|
{"key": "embed_job_id", "value": "", "type": "string"},
|
|
{"key": "poll_retries", "value": "0", "type": "string"},
|
|
{"key": "embed_poll_retries", "value": "0", "type": "string"}
|
|
],
|
|
"item": [
|
|
{
|
|
"name": "Submit Chat Completion",
|
|
"event": [
|
|
{
|
|
"listen": "prerequest",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"pm.collectionVariables.set('poll_retries', '0');"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"listen": "test",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"var code = pm.response.code;",
|
|
"if (code === 404) {",
|
|
" pm.test('Async not configured (404)', function() { pm.expect(true).to.be.true; });",
|
|
" if (typeof postman !== 'undefined' && postman.setNextRequest) {",
|
|
" postman.setNextRequest('Submit with stream (expect 400)');",
|
|
" }",
|
|
" return;",
|
|
"}",
|
|
"pm.test('Status 202', function() { pm.expect(code).to.equal(202); });",
|
|
"if (code === 202) {",
|
|
" var json = pm.response.json();",
|
|
" pm.test('Has job_id', function() { pm.expect(json.id).to.be.a('string'); });",
|
|
" pm.test('Status is pending', function() { pm.expect(json.status).to.equal('pending'); });",
|
|
" pm.collectionVariables.set('job_id', json.id);",
|
|
"}"
|
|
]
|
|
}
|
|
}
|
|
],
|
|
"request": {
|
|
"method": "POST",
|
|
"header": [{"key": "Content-Type", "value": "application/json"}],
|
|
"body": {
|
|
"mode": "raw",
|
|
"raw": "{\n \"model\": \"{{provider}}/{{chat_model}}\",\n \"messages\": [{\"role\": \"user\", \"content\": \"Hello\"}],\n \"max_completion_tokens\": 10,\n \"stream\": false\n}"
|
|
},
|
|
"url": {
|
|
"raw": "{{base_url}}/v1/async/chat/completions",
|
|
"host": ["{{base_url}}"],
|
|
"path": ["v1", "async", "chat", "completions"]
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"name": "Poll Chat Completion",
|
|
"event": [
|
|
{
|
|
"listen": "test",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"var code = pm.response.code;",
|
|
"if (code === 404) { pm.test('Skip - async not configured', function() { pm.expect(true).to.be.true; }); return; }",
|
|
"var json = pm.response.json();",
|
|
"var status = json.status;",
|
|
"if (status === 'pending' || status === 'processing') {",
|
|
" pm.test('HTTP 202 while pending/processing', function() { pm.expect(code).to.equal(202); });",
|
|
" var retries = parseInt(pm.collectionVariables.get('poll_retries') || '0', 10);",
|
|
" if (retries < 10) {",
|
|
" pm.collectionVariables.set('poll_retries', String(retries + 1));",
|
|
" var end = Date.now() + 3000;",
|
|
" while (Date.now() < end) {}",
|
|
" if (typeof postman !== 'undefined' && postman.setNextRequest) {",
|
|
" postman.setNextRequest('Poll Chat Completion');",
|
|
" }",
|
|
" } else {",
|
|
" pm.test('Job completed or max retries', function() { pm.expect(status).to.be.oneOf(['completed', 'failed']); });",
|
|
" }",
|
|
"} else if (status === 'completed') {",
|
|
" pm.test('HTTP 200 when completed', function() { pm.expect(code).to.equal(200); });",
|
|
" pm.test('Job completed', function() { pm.expect(status).to.equal('completed'); });",
|
|
" pm.test('Has result', function() { pm.expect(json.result).to.not.be.undefined; });",
|
|
" pm.test('Result has choices with content', function() {",
|
|
" pm.expect(json.result).to.have.property('choices').that.is.an('array').and.has.length.above(0);",
|
|
" pm.expect(json.result.choices[0]).to.have.property('message');",
|
|
" pm.expect(json.result.choices[0].message).to.have.property('content').that.is.a('string');",
|
|
" });",
|
|
"} else if (status === 'failed') {",
|
|
" pm.test('HTTP 200 when failed', function() { pm.expect(code).to.equal(200); });",
|
|
" pm.test('Job failed (acceptable)', function() { pm.expect(status).to.equal('failed'); });",
|
|
"}"
|
|
]
|
|
}
|
|
}
|
|
],
|
|
"request": {
|
|
"method": "GET",
|
|
"header": [],
|
|
"url": {
|
|
"raw": "{{base_url}}/v1/async/chat/completions/{{job_id}}",
|
|
"host": ["{{base_url}}"],
|
|
"path": ["v1", "async", "chat", "completions", "{{job_id}}"]
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"name": "Submit Embedding with TTL",
|
|
"event": [
|
|
{
|
|
"listen": "prerequest",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"pm.collectionVariables.set('embed_poll_retries', '0');"
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"listen": "test",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"var code = pm.response.code;",
|
|
"if (code === 404) { pm.test('Skip - async not configured', function() { pm.expect(true).to.be.true; }); return; }",
|
|
"pm.test('Status 202', function() { pm.expect(code).to.equal(202); });",
|
|
"if (code === 202) {",
|
|
" var json = pm.response.json();",
|
|
" pm.collectionVariables.set('embed_job_id', json.id);",
|
|
" pm.test('expires_at is set when TTL provided', function() { pm.expect(json.expires_at).to.not.be.undefined; });",
|
|
"}"
|
|
]
|
|
}
|
|
}
|
|
],
|
|
"request": {
|
|
"method": "POST",
|
|
"header": [
|
|
{"key": "Content-Type", "value": "application/json"},
|
|
{"key": "x-bf-async-job-result-ttl", "value": "60"}
|
|
],
|
|
"body": {
|
|
"mode": "raw",
|
|
"raw": "{\n \"model\": \"{{provider}}/{{embedding_model}}\",\n \"input\": \"Hello world\",\n \"encoding_format\": \"float\"\n}"
|
|
},
|
|
"url": {
|
|
"raw": "{{base_url}}/v1/async/embeddings",
|
|
"host": ["{{base_url}}"],
|
|
"path": ["v1", "async", "embeddings"]
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"name": "Poll Embedding",
|
|
"event": [
|
|
{
|
|
"listen": "test",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"var code = pm.response.code;",
|
|
"if (code === 404) { pm.test('Skip - async not configured', function() { pm.expect(true).to.be.true; }); return; }",
|
|
"var json = pm.response.json();",
|
|
"var status = json.status;",
|
|
"if (status === 'pending' || status === 'processing') {",
|
|
" pm.test('HTTP 202 while pending/processing', function() { pm.expect(code).to.equal(202); });",
|
|
" var retries = parseInt(pm.collectionVariables.get('embed_poll_retries') || '0', 10);",
|
|
" if (retries < 10) {",
|
|
" pm.collectionVariables.set('embed_poll_retries', String(retries + 1));",
|
|
" var end = Date.now() + 3000;",
|
|
" while (Date.now() < end) {}",
|
|
" if (typeof postman !== 'undefined' && postman.setNextRequest) {",
|
|
" postman.setNextRequest('Poll Embedding');",
|
|
" }",
|
|
" }",
|
|
"} else if (status === 'completed') {",
|
|
" pm.test('HTTP 200 when completed', function() { pm.expect(code).to.equal(200); });",
|
|
" pm.test('Embedding job completed', function() { pm.expect(status).to.equal('completed'); });",
|
|
" pm.test('Result has embedding data', function() {",
|
|
" pm.expect(json.result).to.not.be.undefined;",
|
|
" pm.expect(json.result).to.have.property('data').that.is.an('array').and.has.length.above(0);",
|
|
" pm.expect(json.result.data[0]).to.have.property('embedding').that.is.an('array');",
|
|
" });",
|
|
"} else if (status === 'failed') {",
|
|
" pm.test('HTTP 200 when failed', function() { pm.expect(code).to.equal(200); });",
|
|
"}"
|
|
]
|
|
}
|
|
}
|
|
],
|
|
"request": {
|
|
"method": "GET",
|
|
"header": [],
|
|
"url": {
|
|
"raw": "{{base_url}}/v1/async/embeddings/{{embed_job_id}}",
|
|
"host": ["{{base_url}}"],
|
|
"path": ["v1", "async", "embeddings", "{{embed_job_id}}"]
|
|
}
|
|
}
|
|
},
|
|
{
|
|
"name": "Submit with stream (expect 400)",
|
|
"event": [
|
|
{
|
|
"listen": "test",
|
|
"script": {
|
|
"type": "text/javascript",
|
|
"exec": [
|
|
"var code = pm.response.code;",
|
|
"if (code === 404) { pm.test('Skip - async not configured', function() { pm.expect(true).to.be.true; }); return; }",
|
|
"pm.test('Streaming not supported on async - expect 400', function() { pm.expect(code).to.equal(400); });"
|
|
]
|
|
}
|
|
}
|
|
],
|
|
"request": {
|
|
"method": "POST",
|
|
"header": [{"key": "Content-Type", "value": "application/json"}],
|
|
"body": {
|
|
"mode": "raw",
|
|
"raw": "{\n \"model\": \"{{provider}}/{{chat_model}}\",\n \"messages\": [{\"role\": \"user\", \"content\": \"Hello\"}],\n \"stream\": true\n}"
|
|
},
|
|
"url": {
|
|
"raw": "{{base_url}}/v1/async/chat/completions",
|
|
"host": ["{{base_url}}"],
|
|
"path": ["v1", "async", "chat", "completions"]
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|