matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-18 11:47:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Sync with upstream ollama/ollama and restore Tesla K80 (compute 3.7) support

Browse Source 
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- Removed outdated CI/CD workflows (tesla-k80-*.yml)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>
This commit is contained in:
Shang Chieh Tseng
2025-11-05 14:03:05 +08:00
parent fabe2c5cb7
commit ef14fb5b26
817 changed files with 241634 additions and 70888 deletions
Download Patch File Download Diff File Expand all files Collapse all files

102
server/prompt_test.go
View File

@@ -27,16 +27,18 @@ func TestChatPrompt(t *testing.T) {
visionModel := Model{Template: tmpl, ProjectorPaths: []string{"vision"}}
cases := []struct {
name string
model Model
limit int
msgs []api.Message
name string
model Model
limit int
truncate bool
msgs []api.Message
expect
}{
{
name: "messages",
model: visionModel,
limit: 64,
name: "messages",
model: visionModel,
limit: 64,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -47,9 +49,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "truncate messages",
model: visionModel,
limit: 1,
name: "truncate messages",
model: visionModel,
limit: 1,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -60,9 +63,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "truncate messages with image",
model: visionModel,
limit: 64,
name: "truncate messages with image",
model: visionModel,
limit: 64,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -76,9 +80,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "truncate messages with images",
model: visionModel,
limit: 64,
name: "truncate messages with images",
model: visionModel,
limit: 64,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!", Images: []api.ImageData{[]byte("something")}},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -92,9 +97,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "messages with images",
model: visionModel,
limit: 2048,
name: "messages with images",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!", Images: []api.ImageData{[]byte("something")}},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -109,9 +115,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "message with image tag",
model: visionModel,
limit: 2048,
name: "message with image tag",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry! [img]", Images: []api.ImageData{[]byte("something")}},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -126,9 +133,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "messages with interleaved images",
model: visionModel,
limit: 2048,
name: "messages with interleaved images",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "user", Images: []api.ImageData{[]byte("something")}},
@@ -145,9 +153,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "truncate message with interleaved images",
model: visionModel,
limit: 1024,
name: "truncate message with interleaved images",
model: visionModel,
limit: 1024,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "user", Images: []api.ImageData{[]byte("something")}},
@@ -163,9 +172,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "message with system prompt",
model: visionModel,
limit: 2048,
name: "message with system prompt",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "system", Content: "You are the Test Who Lived."},
{Role: "user", Content: "You're a test, Harry!"},
@@ -177,9 +187,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "out of order system",
model: visionModel,
limit: 2048,
name: "out of order system",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "assistant", Content: "I-I'm a what?"},
@@ -191,9 +202,10 @@ func TestChatPrompt(t *testing.T) {
},
},
{
name: "multiple images same prompt",
model: visionModel,
limit: 2048,
name: "multiple images same prompt",
model: visionModel,
limit: 2048,
truncate: true,
msgs: []api.Message{
{Role: "user", Content: "Compare these two pictures of hotdogs", Images: []api.ImageData{[]byte("one hotdog"), []byte("two hotdogs")}},
},
@@ -202,6 +214,20 @@ func TestChatPrompt(t *testing.T) {
images: [][]byte{[]byte("one hotdog"), []byte("two hotdogs")},
},
},
{
name: "no truncate with limit exceeded",
model: visionModel,
limit: 10,
truncate: false,
msgs: []api.Message{
{Role: "user", Content: "You're a test, Harry!"},
{Role: "assistant", Content: "I-I'm a what?"},
{Role: "user", Content: "A test. And a thumping good one at that, I'd wager."},
},
expect: expect{
prompt: "You're a test, Harry! I-I'm a what? A test. And a thumping good one at that, I'd wager. ",
},
},
}
for _, tt := range cases {
@@ -209,7 +235,7 @@ func TestChatPrompt(t *testing.T) {
model := tt.model
opts := api.Options{Runner: api.Runner{NumCtx: tt.limit}}
think := false
prompt, images, err := chatPrompt(t.Context(), &model, mockRunner{}.Tokenize, &opts, tt.msgs, nil, &api.ThinkValue{Value: think})
prompt, images, err := chatPrompt(t.Context(), &model, mockRunner{}.Tokenize, &opts, tt.msgs, nil, &api.ThinkValue{Value: think}, tt.truncate)
if tt.error == nil && err != nil {
t.Fatal(err)
} else if tt.error != nil && err != tt.error {