Availability of 0.2.1

Author: cool-japan

CHANGELOG.md

@@ -5,6 +5,23 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.1] - 2026-06-06
+
+### Changed
+
+- **Raised compile optimization for the `test` and `dev` profiles** (`Cargo.toml`): added `[profile.test] opt-level = 2` and `[profile.dev.package."*"] opt-level = 3`, so test binaries and all dependencies (including workspace path-deps like `oxibonsai-model` / `oxibonsai-kernels` when built as deps of another crate's tests) are optimized and autovectorized. At the default `opt-level = 0` the workspace's tests run real numeric work unoptimized — the parity golden references and model forward passes took *minutes* (e.g. the DiT-shape joint-attention CPU reference ~14.5 GFLOP, and the speculative decoder's ~240 forward passes over the 151936-row vocab). The crate under active edit stays at `dev` opt-level 0, so incremental compiles of your own code remain fast. Float results are unchanged: Rust does not enable fast-math, so `opt-level` does not reassociate reductions — parity gates (cos≥0.999) stay bit-stable.
+- **Bumped `oxiarc-deflate` 0.3.2 → 0.3.3** (`Cargo.toml`, `[workspace.dependencies]`): tracks the latest OxiARC release per the COOLJAPAN Latest-crates policy. `oxiarc-deflate` provides the Pure-Rust DEFLATE backend for PNG output in `oxibonsai-image`; the substantive fixes in the 0.3.3 OxiARC release land in sibling crates (`oxiarc-brotli` high-entropy round-trip), so for OxiBonsai this is a version-tracking bump with no change to DEFLATE/PNG behavior.
+
+### Fixed
+
+- **VAE precheck rejected a valid `.safetensors` file** (`oxibonsai-image`, `src/pipeline.rs`): the text-to-image precheck used `is_dir()`, so a valid `.safetensors` FILE passed via `--vae` / `OXI_VAE_WEIGHTS` was rejected with "VAE weights dir not found" before any loading — even though `VaeWeights::open` and the docs both accept a file. The precheck now accepts a file **or** a directory (`is_file() || is_dir()`), error wording is corrected, and the stale doc-comment is fixed. Added regression tests `test_issue_9_*`. (#9)
+
+### Documentation
+
+- **Corrected stale HuggingFace asset paths** (`docs/IMAGEN.md`, `crates/oxibonsai-image/README.md`): the DiT lives under `transformer-packed-mflux/`, and the text encoder + tokenizer ship inside the main `prism-ml/bonsai-image-ternary-4B-mlx-2bit` repo under `text_encoder-mlx-4bit/` (the standalone `prism-ml/text_encoder-mlx-4bit` repo does not exist). Consumer paths updated to match `hf download --local-dir` layout, and the bundled vs. gated `black-forest-labs/FLUX.2-dev` VAE choice is clarified. Verified against the live HF API. (#8)
+
+---
+
 ## [0.2.0] - 2026-06-02
 
 ### Added
@@ -314,6 +331,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Comprehensive test suite (140 tests)
 - Cross-platform support (macOS, Linux, Windows, WASM)
 
+[0.2.1]: https://github.com/cool-japan/oxibonsai/compare/v0.2.0...v0.2.1
 [0.2.0]: https://github.com/cool-japan/oxibonsai/compare/v0.1.5...v0.2.0
 [0.1.2]: https://github.com/cool-japan/oxibonsai/compare/v0.1.1...v0.1.2
 [0.1.1]: https://github.com/cool-japan/oxibonsai/compare/v0.1.0...v0.1.1

Cargo.toml

@@ -14,7 +14,7 @@ members = [
 ]
 
 [workspace.package]
-version = "0.2.0"
+version = "0.2.1"
 edition = "2021"
 license = "Apache-2.0"
 repository = "https://github.com/cool-japan/oxibonsai"
@@ -27,23 +27,23 @@ categories = ["science", "algorithms", "command-line-utilities"]
 
 [workspace.dependencies]
 # ── Internal crates ──
-oxibonsai-core = { path = "crates/oxibonsai-core", version = "0.2.0" }
-oxibonsai-image = { path = "crates/oxibonsai-image", version = "0.2.0" }
-oxibonsai-kernels = { path = "crates/oxibonsai-kernels", version = "0.2.0" }
-oxibonsai-model = { path = "crates/oxibonsai-model", version = "0.2.0" }
-oxibonsai-runtime = { path = "crates/oxibonsai-runtime", version = "0.2.0", default-features = false }
-oxibonsai-tokenizer = { path = "crates/oxibonsai-tokenizer", version = "0.2.0" }
-oxibonsai-serve = { path = "crates/oxibonsai-serve", version = "0.2.0" }
-oxibonsai-rag = { path = "crates/oxibonsai-rag", version = "0.2.0" }
-oxibonsai-eval = { path = "crates/oxibonsai-eval", version = "0.2.0" }
-oxibonsai = { path = "crates/oxibonsai", version = "0.2.0" }
+oxibonsai-core = { path = "crates/oxibonsai-core", version = "0.2.1" }
+oxibonsai-image = { path = "crates/oxibonsai-image", version = "0.2.1" }
+oxibonsai-kernels = { path = "crates/oxibonsai-kernels", version = "0.2.1" }
+oxibonsai-model = { path = "crates/oxibonsai-model", version = "0.2.1" }
+oxibonsai-runtime = { path = "crates/oxibonsai-runtime", version = "0.2.1", default-features = false }
+oxibonsai-tokenizer = { path = "crates/oxibonsai-tokenizer", version = "0.2.1" }
+oxibonsai-serve = { path = "crates/oxibonsai-serve", version = "0.2.1" }
+oxibonsai-rag = { path = "crates/oxibonsai-rag", version = "0.2.1" }
+oxibonsai-eval = { path = "crates/oxibonsai-eval", version = "0.2.1" }
+oxibonsai = { path = "crates/oxibonsai", version = "0.2.1" }
 
 # ── COOLJAPAN Ecosystem (Pure Rust Policy) ──
 scirs2-core = { version = "0.5.0", default-features = false }
 scirs2-linalg = { version = "0.5.0", default-features = false }
 scirs2-neural = { version = "0.5.0", default-features = false }
 oxifft = "0.3.2"
-oxiarc-deflate = "0.3.2"
+oxiarc-deflate = "0.3.3"
 
 # ── GPU backends ──
 metal = "0.33.0"
@@ -229,3 +229,21 @@ harness = false
 lto = true
 codegen-units = 1
 opt-level = 3
+
+# Test/dev runtime performance. This workspace's tests run real numeric work
+# (parity golden references, model forward passes). At the default dev/test
+# opt-level=0 those are unoptimized + non-autovectorized, so some tests take
+# *minutes* (e.g. the DiT-shape joint-attention CPU reference ~14.5 GFLOP, and
+# the speculative decoder's ~240 forward passes over the 151936-row vocab).
+# Float results are unchanged: Rust does not enable fast-math, so opt-level does
+# not reassociate reductions — parity gates (cos>=0.999) stay bit-stable.
+[profile.test]
+opt-level = 2
+
+# Dependencies (incl. workspace path-deps like oxibonsai-model/-kernels when
+# built as deps of another crate's tests) compile under the `dev` profile during
+# `cargo test`/`nextest`. Optimize them so the model forward in the runtime tests
+# isn't the bottleneck. The crate you're actively editing stays at dev
+# opt-level=0, so incremental compiles of your own code remain fast.
+[profile.dev.package."*"]
+opt-level = 3

README.md

@@ -17,7 +17,7 @@ To our knowledge, OxiBonsai is the first pure-Rust — C/C++/Fortran-free, zero-
 
 ## Status
 
-**Version 0.2.0** — released 2026-06-02 · **4,667 tests passing** · ~177k lines of Rust · Pure Rust
+**Version 0.2.1** — released 2026-06-06 · **4,667 tests passing** · ~177k lines of Rust · Pure Rust
 
 | Crate | Status | Tests |
 |-------|--------|-------|
@@ -173,7 +173,7 @@ This installs the `oxibonsai` binary. Rust 1.86+ required.
 
 ```toml
 [dependencies]
-oxibonsai = "0.2.0"
+oxibonsai = "0.2.1"
 ```
 
 ### Build from source (for development)

TODO.md

@@ -1,7 +1,7 @@
 # OxiBonsai TODO
 
 > Pure Rust 1-bit LLM inference engine for PrismML Bonsai models
-> 619 source files, ~177,000 lines of Rust code, 4,667 tests passing across workspace — verified 2026-06-02 (0.2.0)
+> 619 source files, ~177,000 lines of Rust code, 4,667 tests passing across workspace — verified 2026-06-02 (0.2.1)
 
 ## Bonsai-Image Port (FLUX.2 Text-to-Image, Pure Rust) — COMPLETE
 

crates/oxibonsai-core/README.md

@@ -1,6 +1,6 @@
 # oxibonsai-core
 
-[![Version](https://img.shields.io/badge/version-0.2.0-blue.svg)](https://crates.io/crates/oxibonsai-core)
+[![Version](https://img.shields.io/badge/version-0.2.1-blue.svg)](https://crates.io/crates/oxibonsai-core)
 [![Tests](https://img.shields.io/badge/tests-207%20passing-brightgreen.svg)]()
 [![Status](https://img.shields.io/badge/status-stable-brightgreen.svg)]()
 
@@ -37,7 +37,7 @@ Part of the [OxiBonsai](https://github.com/cool-japan/oxibonsai) project.
 
 ```toml
 [dependencies]
-oxibonsai-core = "0.2.0"
+oxibonsai-core = "0.2.1"
 ```
 
 ## License

crates/oxibonsai-core/TODO.md

@@ -2,7 +2,7 @@
 
 > GGUF loader, quant block types, tensor types, model config, error types
 > 16 files, ~5,215 lines, 207 tests
-> Version: 0.2.0 — Last updated: 2026-06-02
+> Version: 0.2.1 — Last updated: 2026-06-06
 
 ## Status: ✅ All Features Complete
 

crates/oxibonsai-eval/README.md

@@ -1,6 +1,6 @@
 # oxibonsai-eval
 
-[![Version](https://img.shields.io/badge/version-0.2.0-blue.svg)](https://crates.io/crates/oxibonsai-eval)
+[![Version](https://img.shields.io/badge/version-0.2.1-blue.svg)](https://crates.io/crates/oxibonsai-eval)
 [![Status](https://img.shields.io/badge/status-stable-brightgreen.svg)]()
 [![Tests](https://img.shields.io/badge/tests-513%20passing-brightgreen.svg)]()
 
@@ -14,7 +14,7 @@ Part of the [OxiBonsai](https://github.com/cool-japan/oxibonsai) project.
 
 ## Status
 
-**Stable** (v0.2.0) — 513 tests passing.
+**Stable** (v0.2.1) — 513 tests passing.
 
 ## Features
 
@@ -35,7 +35,7 @@ Part of the [OxiBonsai](https://github.com/cool-japan/oxibonsai) project.
 
 ```toml
 [dependencies]
-oxibonsai-eval = "0.2.0"
+oxibonsai-eval = "0.2.1"
 ```
 
 ```rust

crates/oxibonsai-eval/TODO.md

@@ -2,8 +2,8 @@
 
 > Model evaluation harness: perplexity, accuracy, ROUGE, throughput metrics
 > 19 source files, 11 test files, 219 integration + proptest tests passing
-> Version: 0.2.0 — Status: Stable
-> Last updated: 2026-06-02
+> Version: 0.2.1 — Status: Stable
+> Last updated: 2026-06-06
 
 ## Status: Stable — Phase 18: WinoGrande + BoolQ evaluators added
 

crates/oxibonsai-image/README.md

@@ -2,7 +2,7 @@
 
 Pure-Rust text-to-image pipeline: FLUX.2-Klein DiT (TQ2_0_g128 ternary) + AutoencoderKLFlux2 VAE + Qwen3-4B 4-bit text encoder + PNG output, all parity-validated against the MLX reference at cosine ≥ 0.999.
 
-**Version:** 0.2.0
+**Version:** 0.2.1
 
 Part of the [OxiBonsai](https://github.com/cool-japan/oxibonsai) project.
 
@@ -59,7 +59,7 @@ Add to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-oxibonsai-image = { version = "0.2.0", features = ["metal"] }  # or "native-cuda"
+oxibonsai-image = { version = "0.2.1", features = ["metal"] }  # or "native-cuda"
 ```
 
 Build with the matching GPU feature:
@@ -91,9 +91,9 @@ let cfg = TextToImageCfg {
     height: 512,
     guidance: 1.0,
     dit_gguf: PathBuf::from("./bonsai-dit.gguf"),
-    vae_weights_dir: PathBuf::from("./flux2/vae/diffusion_pytorch_model.safetensors"),
-    te_source: TeSource::Mlx4bit(PathBuf::from("./bonsai-te/model.safetensors")),
-    tokenizer_dir: PathBuf::from("./bonsai-te"),
+    vae_weights_dir: PathBuf::from("./bonsai-vae/vae/diffusion_pytorch_model.safetensors"),
+    te_source: TeSource::Mlx4bit(PathBuf::from("./bonsai-te/text_encoder-mlx-4bit/model.safetensors")),
+    tokenizer_dir: PathBuf::from("./bonsai-te/text_encoder-mlx-4bit"),
     golden_override: None,
 };
 
@@ -119,14 +119,14 @@ flags take precedence). Create a `.env` in your working directory:
 OXI_DIT_GGUF=./bonsai-dit.gguf
 
 # Text encoder: 4-bit MLX model.safetensors (≈2.1 GB)
-OXI_TE_4BIT=./bonsai-te/model.safetensors
+OXI_TE_4BIT=./bonsai-te/text_encoder-mlx-4bit/model.safetensors
 
 # Tokenizer directory containing tokenizer.json
 # (defaults to the TE .safetensors parent when omitted)
-OXI_TE_TOKENIZER_DIR=./bonsai-te
+OXI_TE_TOKENIZER_DIR=./bonsai-te/text_encoder-mlx-4bit
 
 # VAE weights: .safetensors file or legacy .npy directory
-OXI_VAE_WEIGHTS=./flux2/vae/diffusion_pytorch_model.safetensors
+OXI_VAE_WEIGHTS=./bonsai-vae/vae/diffusion_pytorch_model.safetensors
 ```
 
 **GPU stage toggles** (default on; set to `"0"` to opt out):
@@ -149,18 +149,24 @@ You need three model assets plus a tokenizer. Downloads use the HuggingFace CLI
 
 ```bash
 # 1. DiT: download ternary MLX checkpoint and convert to GGUF
+#    (hf download keeps the repo subfolder, so files land under ./bonsai-dit/transformer-packed-mflux/)
 hf download prism-ml/bonsai-image-ternary-4B-mlx-2bit \
-    diffusion_pytorch_model.safetensors --local-dir ./bonsai-dit
+    transformer-packed-mflux/diffusion_pytorch_model.safetensors --local-dir ./bonsai-dit
 cargo run -p oxibonsai-model --example mlx_image_convert --release -- \
-    ./bonsai-dit/diffusion_pytorch_model.safetensors ./bonsai-dit.gguf tq2_0_g128
+    ./bonsai-dit/transformer-packed-mflux/diffusion_pytorch_model.safetensors ./bonsai-dit.gguf tq2_0_g128
 
-# 2. Text encoder (no conversion — native 4-bit Rust loader)
-hf download prism-ml/text_encoder-mlx-4bit \
-    model.safetensors tokenizer.json --local-dir ./bonsai-te
+# 2. Text encoder + tokenizer (same repo; no conversion — native 4-bit Rust loader)
+hf download prism-ml/bonsai-image-ternary-4B-mlx-2bit \
+    text_encoder-mlx-4bit/model.safetensors text_encoder-mlx-4bit/tokenizer.json \
+    --local-dir ./bonsai-te
 
-# 3. VAE (no conversion — native safetensors Rust loader)
-hf download black-forest-labs/FLUX.2-dev \
-    vae/diffusion_pytorch_model.safetensors --local-dir ./flux2
+# 3. VAE (no conversion — native safetensors Rust loader).
+#    Option A (simplest, non-gated): the VAE bundled in the same PrismML repo
+hf download prism-ml/bonsai-image-ternary-4B-mlx-2bit \
+    vae/diffusion_pytorch_model.safetensors --local-dir ./bonsai-vae
+#    Option B (canonical, gated — needs huggingface-cli login + license):
+# hf download black-forest-labs/FLUX.2-dev \
+#     vae/diffusion_pytorch_model.safetensors --local-dir ./flux2
 ```
 
 For the full step-by-step walkthrough see [`docs/IMAGEN.md`](../../docs/IMAGEN.md).

crates/oxibonsai-image/TODO.md

@@ -1,6 +1,6 @@
 # oxibonsai-image TODO
 
-> v0.2.0 — 2026-06-02
+> v0.2.1 — 2026-06-06
 > **STABLE** — imagen pipeline complete, GPU-accelerated, parity-validated.
 
 ## Completed