Files
ollama37/integration/basic_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- Removed outdated CI/CD workflows (tesla-k80-*.yml)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

147 lines
3.7 KiB
Go

//go:build integration
package integration
import (
"context"
"log/slog"
"os"
"runtime"
"testing"
"time"
"github.com/ollama/ollama/api"
)
func TestBlueSky(t *testing.T) {
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
// Set up the test data
req := api.ChatRequest{
Model: smol,
Messages: []api.Message{
{
Role: "user",
Content: blueSkyPrompt,
},
},
Stream: &stream,
Options: map[string]any{
"temperature": 0,
"seed": 123,
},
}
ChatTestHelper(ctx, t, req, blueSkyExpected)
}
func TestUnicode(t *testing.T) {
skipUnderMinVRAM(t, 6)
ctx, cancel := context.WithTimeout(context.Background(), 3*time.Minute)
defer cancel()
// Set up the test data
req := api.ChatRequest{
// DeepSeek has a Unicode tokenizer regex, making it a unicode torture test
Model: "deepseek-coder-v2:16b-lite-instruct-q2_K", // TODO is there an ollama-engine model we can switch to and keep the coverage?
Messages: []api.Message{
{
Role: "user",
Content: "天空为什么是蓝色的?", // Why is the sky blue?
},
},
Stream: &stream,
Options: map[string]any{
"temperature": 0,
"seed": 123,
// Workaround deepseek context shifting bug
"num_ctx": 8192,
"num_predict": 2048,
},
}
client, _, cleanup := InitServerConnection(ctx, t)
defer cleanup()
if err := PullIfMissing(ctx, client, req.Model); err != nil {
t.Fatal(err)
}
slog.Info("loading", "model", req.Model)
err := client.Generate(ctx, &api.GenerateRequest{Model: req.Model}, func(response api.GenerateResponse) error { return nil })
if err != nil {
t.Fatalf("failed to load model %s: %s", req.Model, err)
}
defer func() {
// best effort unload once we're done with the model
client.Generate(ctx, &api.GenerateRequest{Model: req.Model, KeepAlive: &api.Duration{Duration: 0}}, func(rsp api.GenerateResponse) error { return nil })
}()
skipIfNotGPULoaded(ctx, t, client, req.Model, 100)
DoChat(ctx, t, client, req, []string{
"散射", // scattering
"频率", // frequency
}, 120*time.Second, 120*time.Second)
}
func TestExtendedUnicodeOutput(t *testing.T) {
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
// Set up the test data
req := api.ChatRequest{
Model: "gemma2:2b",
Messages: []api.Message{
{
Role: "user",
Content: "Output some smily face emoji",
},
},
Stream: &stream,
Options: map[string]any{
"temperature": 0,
"seed": 123,
},
}
client, _, cleanup := InitServerConnection(ctx, t)
defer cleanup()
if err := PullIfMissing(ctx, client, req.Model); err != nil {
t.Fatal(err)
}
DoChat(ctx, t, client, req, []string{"😀", "😊", "😁", "😂", "😄", "😃"}, 120*time.Second, 120*time.Second)
}
func TestUnicodeModelDir(t *testing.T) {
// This is only useful for Windows with utf-16 characters, so skip this test for other platforms
if runtime.GOOS != "windows" {
t.Skip("Unicode test only applicable to windows")
}
// Only works for local testing
if os.Getenv("OLLAMA_TEST_EXISTING") != "" {
t.Skip("TestUnicodeModelDir only works for local testing, skipping")
}
modelDir, err := os.MkdirTemp("", "ollama_埃")
if err != nil {
t.Fatal(err)
}
defer os.RemoveAll(modelDir)
slog.Info("unicode", "OLLAMA_MODELS", modelDir)
t.Setenv("OLLAMA_MODELS", modelDir)
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
req := api.ChatRequest{
Model: smol,
Messages: []api.Message{
{
Role: "user",
Content: blueSkyPrompt,
},
},
Stream: &stream,
Options: map[string]any{
"temperature": 0,
"seed": 123,
},
}
ChatTestHelper(ctx, t, req, blueSkyExpected)
}