core/go-mlx
8
0
Fork 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions
46 commits 1 branch 4 tags 1.2 MiB
Commit graph

9 commits

Author SHA1 Message Date
Snider
443347a2f8 fix(metal): address 4 minor code review items 
- Rename New() → newArray() to signal internal-only intent (112 usages)
- Remove unused Collect() function and its test
- Fix discarded json.Unmarshal error in qwen3.go
- Document AsStrided stride formula in gemma3.go

Co-Authored-By: Virgil <virgil@lethean.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-02-19 21:36:40 +00:00
Snider
754d6e2f93 fix(metal): error handling audit — propagate MLX errors instead of swallowing 
Replace checkError() log+swallow with lastError() that returns real MLX
error messages. Add Eval/EvalAsync as error-returning variants of
Materialize. Generate loop now propagates GPU errors via model.Err().
LoadAllSafetensors returns (map, error). Model loaders check lastError()
after safetensors load. 180 tests passing.

Co-Authored-By: Virgil <virgil@lethean.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-02-19 20:59:37 +00:00
Snider
ff01175a62 bench(metal): add 29 benchmarks baselined on M3 Ultra 
MatMul (128² to 4096², token projection), Softmax, element-wise
ops, fused Metal kernels (RMSNorm, LayerNorm, RoPE, SDPA), Linear,
Embedding, reductions, and full sampler chain. CGO floor ~170μs.

Co-Authored-By: Virgil <virgil@lethean.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-02-19 20:47:25 +00:00
Snider
7435648f66 docs: update findings and TODO with CLion Claude research results 
Document contiguous fix, TopP/MinP implementations, new bindings.
Update memory management TODO — CLion Claude confirmed safe patterns.
176 tests passing.

Co-Authored-By: Virgil <virgil@lethean.io>
 2026-02-19 20:40:35 +00:00
Snider
bff97ccf19 feat(api): migrate to go-inference shared interfaces 
Replace local TextModel, Backend, Token, Message, and option types with
forge.lthn.ai/core/go-inference. go-mlx is now a pure backend that
registers "metal" into the shared inference registry via init().

Deleted: textmodel.go, options.go, backend.go
Updated: register_metal.go (implements inference.Backend with Available()),
  mlx_test.go (uses inference.* types, 4 new tests), go.mod,
  internal/metal/generate.go (added RepeatPenalty)

159 tests passing (148 internal/metal + 11 root).

Co-Authored-By: Virgil <virgil@lethean.io>
 2026-02-19 20:15:42 +00:00
Snider
95d92fffff docs: update project docs for backend abstraction completion 
- CLAUDE.md: new architecture diagram, public API examples
- TODO.md: Phase 4 marked complete, remaining items noted
- FINDINGS.md: migration completion notes, import cycle resolution

Co-Authored-By: Virgil <virgil@lethean.io>
 2026-02-19 20:07:01 +00:00
Snider
c881813872 docs(design): backend abstraction with internal/metal reorganisation 
Approved design for restructuring go-mlx:
- Root package becomes clean interface (TextModel, LoadModel, Token)
- All CGO code moves to internal/metal/
- Deterministic memory management (Close + per-step cleanup)
- Error propagation instead of silent logging
- mlxlm/ backend placeholder for Python subprocess support

Includes API breaking change communication in FINDINGS.md and
memory management research tasks in cpp/TODO.md.

See: docs/plans/2026-02-19-backend-abstraction-design.md

Co-Authored-By: Virgil <virgil@lethean.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-02-19 19:12:04 +00:00
Snider
37abc496ba test(core): add 86 tests for ops, array, nn, fast kernels 
Phase 1 hardening: cover all previously-untested core operations.

- array_test.go (25): scalar/array creation, shape, clone, free, data access
- ops_test.go (44): arithmetic, math, matmul, reductions, shape ops, indexing, slicing, random
- nn_test.go (8): Linear (dense/bias/LoRA), Embedding, RMSNormModule, RepeatKV
- fast_test.go (9): RMSNorm, LayerNorm, RoPE, ScaledDotProductAttention

Found: Floats()/DataInt32() return wrong data on non-contiguous arrays
(transpose, broadcast, slice views). Documented in FINDINGS.md.

Also: cpp/ workspace docs for CLion Claude session, Go 1.26 impact
assessment, verified go generate → test round-trip (29→115 tests).

Co-Authored-By: Virgil <virgil@lethean.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
 2026-02-19 18:37:30 +00:00
Snider
cae7ef05e8 feat: extract go-mlx from go-ai as standalone Metal inference package 
Split mlx/ directory from forge.lthn.ai/core/go-ai into its own module.
Rewrites import paths, adds CLAUDE.md/TODO.md/FINDINGS.md for dedicated
Claude sessions. Zero external Go deps — pure CGO + mlx-c v0.4.1.

Co-Authored-By: Virgil <virgil@lethean.io>
 2026-02-19 17:57:37 +00:00