klinsmaya/CLIProxyAPI
1
0
Fork 0
mirror of https://gh-proxy.org/https://github.com/klinsmaya/CLIProxyAPI.git synced 2026-11-09 09:56:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(translator): add support for cache_write_tokens in Codex response usage details

Browse Source 
- Introduced handling for `cache_write_tokens` in `input_tokens_details` and mapped to `cached_creation_tokens` in output.
- Added comprehensive tests to validate correct parsing, preservation, and omission of `cache_write_tokens` in both streaming and non-streaming responses.

Closes: #4197
This commit is contained in:
Luis Pater
2026-07-10 23:03:25 +08:00
parent 3554b63721
commit abb5224880
2 changed files with 107 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
internal/translator/codex/openai/chat-completions/codex_openai_response.go
+6
View File
@@ -109,6 +109,9 @@ func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalR
if cachedTokensResult := usageResult.Get("input_tokens_details.cached_tokens"); cachedTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.prompt_tokens_details.cached_tokens", cachedTokensResult.Int())
}
if cacheWriteTokensResult := usageResult.Get("input_tokens_details.cache_write_tokens"); cacheWriteTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.prompt_tokens_details.cached_creation_tokens", cacheWriteTokensResult.Int())
}
if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
}
@@ -357,6 +360,9 @@ func ConvertCodexResponseToOpenAINonStream(_ context.Context, _ string, original
if cachedTokensResult := usageResult.Get("input_tokens_details.cached_tokens"); cachedTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.prompt_tokens_details.cached_tokens", cachedTokensResult.Int())
}
if cacheWriteTokensResult := usageResult.Get("input_tokens_details.cache_write_tokens"); cacheWriteTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.prompt_tokens_details.cached_creation_tokens", cacheWriteTokensResult.Int())
}
if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
template, _ = sjson.SetBytes(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
}
internal/translator/codex/openai/chat-completions/codex_openai_response_test.go
+101
View File
@@ -150,6 +150,107 @@ func TestConvertCodexResponseToOpenAI_NonStreamImageGenerationCallAddsMessageIma
}
}
func TestConvertCodexResponseToOpenAI_StreamForwardsCacheWriteTokens(t *testing.T) {
ctx := context.Background()
var param any
// Seed response.created so response.completed can reuse response metadata.
_ = ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, []byte(`data: {"type":"response.created","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4"}}`), &param)
chunk := []byte(`data: {"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":40},"output_tokens_details":{"reasoning_tokens":5}}}}`)
out := ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, chunk, &param)
if len(out) != 1 {
t.Fatalf("expected 1 chunk, got %d", len(out))
}
assertUsageMapping(t, out[0], 40, true)
}
func TestConvertCodexResponseToOpenAI_StreamOmitsMissingCacheWriteTokens(t *testing.T) {
ctx := context.Background()
var param any
_ = ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, []byte(`data: {"type":"response.created","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4"}}`), &param)
chunk := []byte(`data: {"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30},"output_tokens_details":{"reasoning_tokens":5}}}}`)
out := ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, chunk, &param)
if len(out) != 1 {
t.Fatalf("expected 1 chunk, got %d", len(out))
}
assertUsageMapping(t, out[0], 0, false)
}
func TestConvertCodexResponseToOpenAI_StreamPreservesExplicitZeroCacheWriteTokens(t *testing.T) {
ctx := context.Background()
var param any
_ = ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, []byte(`data: {"type":"response.created","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4"}}`), &param)
chunk := []byte(`data: {"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":0},"output_tokens_details":{"reasoning_tokens":5}}}}`)
out := ConvertCodexResponseToOpenAI(ctx, "gpt-5.4", nil, nil, chunk, &param)
if len(out) != 1 {
t.Fatalf("expected 1 chunk, got %d", len(out))
}
assertUsageMapping(t, out[0], 0, true)
}
func TestConvertCodexResponseToOpenAI_NonStreamForwardsCacheWriteTokens(t *testing.T) {
ctx := context.Background()
raw := []byte(`{"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","status":"completed","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":40},"output_tokens_details":{"reasoning_tokens":5}},"output":[{"type":"message","content":[{"type":"output_text","text":"ok"}]}]}}`)
out := ConvertCodexResponseToOpenAINonStream(ctx, "gpt-5.4", nil, nil, raw, nil)
assertUsageMapping(t, out, 40, true)
}
func TestConvertCodexResponseToOpenAI_NonStreamOmitsMissingCacheWriteTokens(t *testing.T) {
ctx := context.Background()
raw := []byte(`{"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","status":"completed","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30},"output_tokens_details":{"reasoning_tokens":5}},"output":[{"type":"message","content":[{"type":"output_text","text":"ok"}]}]}}`)
out := ConvertCodexResponseToOpenAINonStream(ctx, "gpt-5.4", nil, nil, raw, nil)
assertUsageMapping(t, out, 0, false)
}
func TestConvertCodexResponseToOpenAI_NonStreamPreservesExplicitZeroCacheWriteTokens(t *testing.T) {
ctx := context.Background()
raw := []byte(`{"type":"response.completed","response":{"id":"resp_123","created_at":1700000000,"model":"gpt-5.4","status":"completed","usage":{"input_tokens":100,"output_tokens":20,"total_tokens":120,"input_tokens_details":{"cached_tokens":30,"cache_write_tokens":0},"output_tokens_details":{"reasoning_tokens":5}},"output":[{"type":"message","content":[{"type":"output_text","text":"ok"}]}]}}`)
out := ConvertCodexResponseToOpenAINonStream(ctx, "gpt-5.4", nil, nil, raw, nil)
assertUsageMapping(t, out, 0, true)
}
func assertUsageMapping(t *testing.T, payload []byte, wantCachedCreation int64, expectCachedCreation bool) {
t.Helper()
if got := gjson.GetBytes(payload, "usage.prompt_tokens").Int(); got != 100 {
t.Fatalf("expected prompt_tokens=100, got %d; payload=%s", got, string(payload))
}
if got := gjson.GetBytes(payload, "usage.completion_tokens").Int(); got != 20 {
t.Fatalf("expected completion_tokens=20, got %d; payload=%s", got, string(payload))
}
if got := gjson.GetBytes(payload, "usage.total_tokens").Int(); got != 120 {
t.Fatalf("expected total_tokens=120, got %d; payload=%s", got, string(payload))
}
if got := gjson.GetBytes(payload, "usage.prompt_tokens_details.cached_tokens").Int(); got != 30 {
t.Fatalf("expected cached_tokens=30, got %d; payload=%s", got, string(payload))
}
if got := gjson.GetBytes(payload, "usage.completion_tokens_details.reasoning_tokens").Int(); got != 5 {
t.Fatalf("expected reasoning_tokens=5, got %d; payload=%s", got, string(payload))
}
gotCachedCreation := gjson.GetBytes(payload, "usage.prompt_tokens_details.cached_creation_tokens")
if expectCachedCreation {
if !gotCachedCreation.Exists() {
t.Fatalf("expected cached_creation_tokens to exist, payload=%s", string(payload))
}
if gotCachedCreation.Int() != wantCachedCreation {
t.Fatalf("expected cached_creation_tokens=%d, got %d; payload=%s", wantCachedCreation, gotCachedCreation.Int(), string(payload))
}
return
}
if gotCachedCreation.Exists() {
t.Fatalf("expected cached_creation_tokens to be omitted, payload=%s", string(payload))
}
}
func TestConvertCodexResponseToOpenAI_NonStreamMultiMessageEmptyTrailingKeepsContent(t *testing.T) {
ctx := context.Background()
raw := []byte(`{"type":"response.completed","response":{"id":"resp_1","created_at":1700000000,"model":"gpt-5.5","status":"completed","usage":{"input_tokens":10,"output_tokens":5,"total_tokens":15},"output":[` +