Files
ollama37/openai/openai_test.go
Shang Chieh Tseng ef14fb5b26 Sync with upstream ollama/ollama and restore Tesla K80 (compute 3.7) support
This commit represents a complete rework after pulling the latest changes from
official ollama/ollama repository and re-applying Tesla K80 compatibility patches.

## Key Changes

### CUDA Compute Capability 3.7 Support (Tesla K80)
- Added sm_37 (compute 3.7) to CMAKE_CUDA_ARCHITECTURES in CMakeLists.txt
- Updated CMakePresets.json to include compute 3.7 in "CUDA 11" preset
- Using 37-virtual (PTX with JIT compilation) for maximum compatibility

### Legacy Toolchain Compatibility
- **NVIDIA Driver**: 470.256.02 (last version supporting Kepler/K80)
- **CUDA Version**: 11.4.4 (last CUDA 11.x supporting compute 3.7)
- **GCC Version**: 10.5.0 (required by CUDA 11.4 host_config.h)

### CPU Architecture Trade-offs
Due to GCC 10.5 limitation, sacrificed newer CPU optimizations:
- Alderlake CPU variant enabled WITHOUT AVX_VNNI (requires GCC 11+)
- Still supports: SSE4.2, AVX, F16C, AVX2, BMI2, FMA
- Performance impact: ~3-7% on newer CPUs (acceptable for K80 compatibility)

### Build System Updates
- Modified ml/backend/ggml/ggml/src/ggml-cuda/CMakeLists.txt for compute 3.7
- Added -Wno-deprecated-gpu-targets flag to suppress warnings
- Updated ml/backend/ggml/ggml/src/CMakeLists.txt for Alderlake without AVX_VNNI

### Upstream Sync
Merged latest llama.cpp changes including:
- Enhanced KV cache management with ISWA and hybrid memory support
- Improved multi-modal support (mtmd framework)
- New model architectures (Gemma3, Llama4, Qwen3, etc.)
- GPU backend improvements for CUDA, Metal, and ROCm
- Updated quantization support and GGUF format handling

### Documentation
- Updated CLAUDE.md with comprehensive build instructions
- Documented toolchain constraints and CPU architecture trade-offs
- Removed outdated CI/CD workflows (tesla-k80-*.yml)
- Cleaned up temporary development artifacts

## Rationale

This fork maintains Tesla K80 GPU support (compute 3.7) which was dropped in
official Ollama due to legacy driver/CUDA requirements. The toolchain constraint
creates a deadlock:
- K80 → Driver 470 → CUDA 11.4 → GCC 10 → No AVX_VNNI

We accept the loss of cutting-edge CPU optimizations to enable running modern
LLMs on legacy but still capable Tesla K80 hardware (12GB VRAM per GPU).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-05 14:03:05 +08:00

151 lines
3.4 KiB
Go

package openai
import (
"encoding/base64"
"testing"
"github.com/ollama/ollama/api"
)
const (
prefix = `data:image/jpeg;base64,`
image = `iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=`
)
func TestFromChatRequest_Basic(t *testing.T) {
req := ChatCompletionRequest{
Model: "test-model",
Messages: []Message{
{Role: "user", Content: "Hello"},
},
}
result, err := FromChatRequest(req)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
if result.Model != "test-model" {
t.Errorf("expected model 'test-model', got %q", result.Model)
}
if len(result.Messages) != 1 {
t.Fatalf("expected 1 message, got %d", len(result.Messages))
}
if result.Messages[0].Role != "user" || result.Messages[0].Content != "Hello" {
t.Errorf("unexpected message: %+v", result.Messages[0])
}
}
func TestFromChatRequest_WithImage(t *testing.T) {
imgData, _ := base64.StdEncoding.DecodeString(image)
req := ChatCompletionRequest{
Model: "test-model",
Messages: []Message{
{
Role: "user",
Content: []any{
map[string]any{"type": "text", "text": "Hello"},
map[string]any{
"type": "image_url",
"image_url": map[string]any{"url": prefix + image},
},
},
},
},
}
result, err := FromChatRequest(req)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
if len(result.Messages) != 2 {
t.Fatalf("expected 2 messages, got %d", len(result.Messages))
}
if result.Messages[0].Content != "Hello" {
t.Errorf("expected first message content 'Hello', got %q", result.Messages[0].Content)
}
if len(result.Messages[1].Images) != 1 {
t.Fatalf("expected 1 image, got %d", len(result.Messages[1].Images))
}
if string(result.Messages[1].Images[0]) != string(imgData) {
t.Error("image data mismatch")
}
}
func TestFromCompleteRequest_Basic(t *testing.T) {
temp := float32(0.8)
req := CompletionRequest{
Model: "test-model",
Prompt: "Hello",
Temperature: &temp,
}
result, err := FromCompleteRequest(req)
if err != nil {
t.Fatalf("unexpected error: %v", err)
}
if result.Model != "test-model" {
t.Errorf("expected model 'test-model', got %q", result.Model)
}
if result.Prompt != "Hello" {
t.Errorf("expected prompt 'Hello', got %q", result.Prompt)
}
if tempVal, ok := result.Options["temperature"].(float32); !ok || tempVal != 0.8 {
t.Errorf("expected temperature 0.8, got %v", result.Options["temperature"])
}
}
func TestToUsage(t *testing.T) {
resp := api.ChatResponse{
Metrics: api.Metrics{
PromptEvalCount: 10,
EvalCount: 20,
},
}
usage := ToUsage(resp)
if usage.PromptTokens != 10 {
t.Errorf("expected PromptTokens 10, got %d", usage.PromptTokens)
}
if usage.CompletionTokens != 20 {
t.Errorf("expected CompletionTokens 20, got %d", usage.CompletionTokens)
}
if usage.TotalTokens != 30 {
t.Errorf("expected TotalTokens 30, got %d", usage.TotalTokens)
}
}
func TestNewError(t *testing.T) {
tests := []struct {
code int
want string
}{
{400, "invalid_request_error"},
{404, "not_found_error"},
{500, "api_error"},
}
for _, tt := range tests {
result := NewError(tt.code, "test message")
if result.Error.Type != tt.want {
t.Errorf("NewError(%d) type = %q, want %q", tt.code, result.Error.Type, tt.want)
}
if result.Error.Message != "test message" {
t.Errorf("NewError(%d) message = %q, want %q", tt.code, result.Error.Message, "test message")
}
}
}