mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-09 23:37:06 +00:00
ef14fb5b26
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with comprehensive build instructions - Documented toolchain constraints and CPU architecture trade-offs - Removed outdated CI/CD workflows (tesla-k80-*.yml) - Cleaned up temporary development artifacts ## Rationale This fork maintains Tesla K80 GPU support (compute 3.7) which was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>
109 lines
2.4 KiB
Go
109 lines
2.4 KiB
Go
package discover
|
|
|
|
import (
|
|
"testing"
|
|
|
|
"github.com/ollama/ollama/app/lifecycle"
|
|
)
|
|
|
|
func init() {
|
|
lifecycle.InitLogging()
|
|
}
|
|
|
|
func TestFilterOverlapByLibrary(t *testing.T) {
|
|
type testcase struct {
|
|
name string
|
|
inp map[string]map[string]map[string]int
|
|
exp []bool
|
|
}
|
|
for _, tc := range []testcase{
|
|
{
|
|
name: "empty",
|
|
inp: map[string]map[string]map[string]int{},
|
|
exp: []bool{}, // needs deletion
|
|
},
|
|
{
|
|
name: "single no overlap",
|
|
inp: map[string]map[string]map[string]int{
|
|
"CUDA": {
|
|
"cuda_v12": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 0,
|
|
},
|
|
},
|
|
},
|
|
exp: []bool{false},
|
|
},
|
|
{
|
|
name: "100% overlap pick 2nd",
|
|
inp: map[string]map[string]map[string]int{
|
|
"CUDA": {
|
|
"cuda_v12": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 0,
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 1,
|
|
},
|
|
"cuda_v13": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 2,
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 3,
|
|
},
|
|
},
|
|
},
|
|
exp: []bool{true, true, false, false},
|
|
},
|
|
{
|
|
name: "100% overlap pick 1st",
|
|
inp: map[string]map[string]map[string]int{
|
|
"CUDA": {
|
|
"cuda_v13": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 0,
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 1,
|
|
},
|
|
"cuda_v12": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 2,
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 3,
|
|
},
|
|
},
|
|
},
|
|
exp: []bool{false, false, true, true},
|
|
},
|
|
{
|
|
name: "partial overlap pick older",
|
|
inp: map[string]map[string]map[string]int{
|
|
"CUDA": {
|
|
"cuda_v13": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 0,
|
|
},
|
|
"cuda_v12": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 1,
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 2,
|
|
},
|
|
},
|
|
},
|
|
exp: []bool{true, false, false},
|
|
},
|
|
{
|
|
name: "no overlap",
|
|
inp: map[string]map[string]map[string]int{
|
|
"CUDA": {
|
|
"cuda_v13": {
|
|
"GPU-d7b00605-c0c8-152d-529d-e03726d5dc52": 0,
|
|
},
|
|
"cuda_v12": {
|
|
"GPU-cd6c3216-03d2-a8eb-8235-2ffbf571712e": 1,
|
|
},
|
|
},
|
|
},
|
|
exp: []bool{false, false},
|
|
},
|
|
} {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
needsDelete := make([]bool, len(tc.exp))
|
|
filterOverlapByLibrary(tc.inp, needsDelete)
|
|
for i, exp := range tc.exp {
|
|
if needsDelete[i] != exp {
|
|
t.Fatalf("expected: %v\ngot: %v", tc.exp, needsDelete)
|
|
}
|
|
}
|
|
})
|
|
}
|
|
}
|