Roadmap

Track active work on GitHub Issues.

Completed (v0.1.0)

• [x] CPU simulator runtime
• [x] OpenCL runtime with embedded kernels
• [x] Core ATen ops and LLM ops (silu, softmax, bmm, layer_norm, embedding)
• [x] Fused GEMM + bias + activation kernel
• [x] JIT pointwise compiler with OpenCL codegen
• [x] torch_tvarant.compiler FX fusion passes
• [x] CI, docs, and release infrastructure

In progress / planned

Priority	Item	Issue
High	Fused attention (QK^T + softmax + PV)	#1
High	KV-cache + decode GEMM (M=1)	#2
High	RMSNorm, RoPE, SwiGLU ops	#3
Medium	Shape-specialized GEMM JIT	#4
Medium	Full transformer block fusion	#7
Long-term	Lower to Tvarant RISC-V ISA	#5
Long-term	int8 / FP16 weight quantization	#6

How to pick up work

1. Comment on an issue to claim it
2. Fork and branch from main
3. Follow Contributing
4. Open a PR linking the issue

Versioning

We use Semantic Versioning:

• Patch — bug fixes, kernel correctness
• Minor — new ops, compiler fusions, backward-compatible features
• Major — breaking API or ABI changes

See CHANGELOG.md.