v0.1.8: Universal Cross-Vendor Hardware Support & LangChain Ecosystem

⚡ Turing Engine v0.1.8

🌐 Universal Cross-Vendor Hardware Support

• AMD ROCm / HIP: Added native hardware auto-discovery and Wave32 (RDNA3/4) / Wave64 (CDNA3/MI300X) Matrix Core tuning heuristics (turing/kernels/rocm_heuristics.py).
• Vulkan Compute (SPIR-V): Cross-vendor, vendor-agnostic GPU acceleration for Intel Arc, AMD APUs, Qualcomm Adreno, and ARM Mali (turing/kernels/vulkan_runtime.py).
• Intel XPU (SYCL / OneAPI): Native support for Intel Arc A770, B580 Battlemage, and Data Center GPU Max 1550.
• Hardware Probe CLI: Added scripts/test_universal_backends.py to inspect and verify hardware acceleration in 1 command.

🦜️🔗 LangChain & LangGraph Integration Ecosystem

• Native In-Tree Export: Exported ChatTuring and TuringLLM directly from turing.integrations.langchain with Subspace pruning (sparsity_ratio=0.57) and SVD INT8 headers.
• Standalone Partner Package: Scaffolding for langchain-turing on PyPI (integrations/langchain/).
• LangGraph Multi-Agent Support: Full agentic state graph deliberation support.

🧪 Verification & Reliability

• 98/98 Automated Tests: 100% test pass rate with strict 0 warnings.
• Live Documentation: Deployed to intutic.github.io/turing.
• Pre-compiled Binary Wheel: Attached macOS Apple Silicon (arm64) wheel asset.

Full Changelog: v0.1.7...v0.1.8

v0.1.7: Master Verified 25-Model Matrix & Exact MoE Sizing

⚡ Turing Engine v0.1.7

🌐 Master Frontier Open-Weight Model Matrix (25 Verified Architectures)

• OpenAI GPT-OSS: Added gpt-oss-120b (117B total / 5.1B active MoE) alongside gpt-oss-20b (3.6B active / runs in 8GB VRAM).
• Google Gemma: Registered gemma-2-27b, gemma-4-31b (Dense), and gemma-4-26b (MoE).
• Moonshot Kimi: Aligned kimi-k3 to 2.8T MoE (104B active, 2M context) and kimi-k2.6 (1.04T MoE).
• Zhipu / Z.ai: Aligned glm-5.3-730b (730B MoE) and glm-5.2 (753B MoE).
• DeepSeek: Full single-GPU support for deepseek-r1-1.5b through 70b and host offloading for deepseek-v4-pro (1.6T MoE).
• Meta AI: Registered llama-4-scout, llama-4-maverick, muse-glimmer-30b, and llama-3.3-70b.
• Alibaba Qwen: Registered qwen-2.5-coder-32b, qwen3-coder-30b, qwen3-coder-480b, qwen-3.8-27b, and qwen3.8-max.

🧪 Verification & Reliability

• 25/25 GPU Passes: All 25 model architectures verified on live GPU forward passes with 0 shape errors.
• 94/94 Automated Tests: 100% test pass rate with strict 0 warnings.
• Pre-compiled Binary Wheels: Attached macOS Apple Silicon (arm64) wheel.

Full Changelog: v0.1.6...v0.1.7

v0.1.6: Frontier 2026 Model Registry, Llama 4, Mistral Large 3, Qwen 3.8 Max

🚀 Turing Engine v0.1.6 Release Notes

• Updated 2026 Frontier & MoE Registry: Registered and verified Meta LLaMA 4 Scout & Maverick, Mistral Large 3, Alibaba Qwen 3.8 Max, Qwen3-Coder-480B, DeepSeek-V4 (Flash & Pro 1.6T), GLM-5.3 (753B MoE), and DeepSeek-R1-0528.
• Overhauled README: Direct 30-second quickstart with instant terminal chat and serving commands for all new model families.
• Pre-Built Binary Wheel: Attached turing_engine-0.1.6-cp311-cp311-macosx_15_0_arm64.whl with full native C++20 AVX2 SIMD acceleration.

Full Changelog: v0.1.5...v0.1.6

v0.1.5: 17 Frontier 2026 Model Architectures & Live GPU Validation

🚀 Turing Engine v0.1.5 Release Notes

• Full 17 Frontier 2026 Model Architectures: Added and verified configurations for GLM-5.3 (753B MoE), DeepSeek-V4 (Flash & Pro 1.6T), Kimi-K3 (2M Ctx), Qwen3-Coder (480B, 80B, 30B), Google Gemma-4 (Dense & MoE), Meta Muse Glimmer-30B, and OpenAI GPT-OSS-20B.
• Live GPU Architecture Validation Suite: Introduced scripts/test_all_gpu_architectures.py verifying GPU layer memory allocation, forward pass execution, and tensor geometry on CUDA and Apple Silicon Metal (17/17 passed).
• Synchronized Documentation: Updated all documentation and CLI reference guides for terminal chat and serving endpoints.

Full Changelog: v0.1.4...v0.1.5

v0.1.4: Real HuggingFace Weights, 2026 Frontier Models & Parametric Benchmarks

🚀 Turing Engine v0.1.4 Release Notes

• 100% Real Pretrained Hugging Face Weight Streaming: Stream real safetensors/PyTorch weights directly into Turing Subspace format.
• 2026 Frontier Models Added: Registered and tested DeepSeek-R1 (1.5B–70B), DeepSeek-V4 (Flash & Pro), GLM-5.3 (753B MoE), Moonshot Kimi-K3, Qwen-3.8-27B, Qwen3-Coder-30B, Google Gemma-4-31B, Meta Muse Glimmer-30B, OpenAI GPT-OSS-20B, and Meta LLaMA-3.3-70B.
• Parametric Memory Geometry: Exact analytical formulas dynamically computed from model layer dimensions.
• Physical GPU & Apple Silicon Metal Verification: 1.99x live layer speedup and -96.7% KV memory compression tested on NVIDIA L4 GPU.

Full Changelog: v0.1.3...v0.1.4

v0.1.3

Full Changelog: v0.1.2...v0.1.3

v0.1.2

Full Changelog: v0.1.1...v0.1.2

v0.1.1

Full Changelog: v0.1.0...v0.1.1

Full Changelog: v0.1.0...v0.1.1

Full Changelog: v0.1.0...v0.1.1

v0.1.0

Full Changelog: https://github.com/intutic/turing/commits/v0.1.0

Full Changelog: https://github.com/intutic/turing/commits/v0.1.0

Full Changelog: https://github.com/intutic/turing/commits/v0.1.0

Full Changelog: https://github.com/intutic/turing/commits/v0.1.0