Commit Graph
318 Commits
Author SHA1 Message Date
leejet 2f886889e6 refactor: align PixArt weights with upstream layout (#2061) 2026-09-26 02:03:05 +08:00
leejet 19bbbca1c7 refactor: define VAE tile dimensions in image pixels (#2059) 2026-09-25 18:20:32 +08:00
losewayy 39ada0863b feat: add PixArt model family support (#2047) 2026-09-25 18:18:55 +08:00
leejet 4c3cf7543d fix: load safetensors index shards without recursion (#2058) 2026-09-25 16:40:20 +08:00
leejet 510bccf330 fix: map Qwen Image 2.1 LoRAs to fused MLP weights (#2057) 2026-09-25 16:30:34 +08:00
leejet 0a9340c599 fix: scale Qwen Image 2.1 VAE convolutions (#2054) 2026-09-25 15:52:35 +08:00
leejet b167b942f7 fix: align Qwen Image 2.1 flow schedule with official defaults (#2048) 2026-09-25 02:43:10 +08:00
leejet 1a2330de68 fix: reserve 128 MiB headroom when selecting monolithic execution (#2046) 2026-09-25 01:53:11 +08:00
leejet 740c7ae193 feat: add configurable Qwen cache types and early cache scheduling (#2045) 2026-09-25 01:51:01 +08:00
Nick Beerbower 4dfe8f5d45 feat: add a stand-alone upscale endpoint to the server (#2026) 2026-09-25 01:13:23 +08:00
stduhpf caa111adf3 feat: optimize cfg special cases with guidance schdeule (#2033) 2026-09-25 01:10:22 +08:00
leejet 88411ef1e0 refactor: centralize circular RoPE and extend image model support (#2039) 2026-09-24 01:45:14 +08:00
Dong Wang 500ef5fa7c fix: map mmapped weights through Metal buffers instead of CPU buffers (#2037) 2026-09-24 00:53:49 +08:00
Dong Wang 70c1dbc01e perf: run one-frame Wan VAE convolutions as 2D convolutions (#2038) 2026-09-24 00:23:31 +08:00
assouan 36746936c0 fix: add graph cuts for MiniMax-H3 text conditioning (#1900) 2026-09-24 00:03:29 +08:00
leejet 2dc7f5408a feat: add Qwen Image 2.1 prefix KV cache (#2035) 2026-09-23 23:08:21 +08:00
leejet e6281b6318 feat: add configurable conditioning cache for all models (#2034) 2026-09-23 22:25:42 +08:00
stduhpf 241518b35d feat: add latent2rgba preview for Qwen-Image 2.1 (#2032) 2026-09-23 22:16:55 +08:00
leejet c92d73c408 fix: preserve alpha when upscaling RGBA images with ESRGAN (#2029) 2026-09-23 02:23:10 +08:00
leejet 28b454bda1 feat: add configurable image input preprocessing (#2028) 2026-09-23 02:12:49 +08:00
LED-M 2bb72947cb perf: cache MiniMax H3 text conditioning (#1966) 2026-09-22 22:19:39 +08:00
Weiqi Gao e112ab5a50 fix: add alpha channel input for Qwen Image 2.1 and relative docs (#2021) 2026-09-22 21:45:06 +08:00
leejet 6dcb5bbd42 fix: handle GPU memory reports and LLM encoding failures (#2020) 2026-09-22 00:28:52 +08:00
leejet 97d932b8f8 fix: restrict VAE tiling retries to allocation failures (#2019) 2026-09-21 23:40:05 +08:00
leejet 74988b290e fix: reject video models in image generation (#2017) 2026-09-21 21:48:42 +08:00
leejet b56c68617d fix: avoid narrowing conversion in SigVQ patch embedding and format code 2026-09-21 01:27:31 +08:00
leejet 187b2561ea feat: add native CUDA SageAttention support (#2005) 2026-09-21 01:16:59 +08:00
fszontaghandleejet 15f335daa5 feat: add LLaDA-Image support (#1968)
Co-authored-by: leejet <leejet714@gmail.com>
2026-09-21 01:14:20 +08:00
Wagner Bruna b8248a869c fix: avoid passing ggml logs as format strings (#2002) 2026-09-21 00:50:52 +08:00
Wagner Bruna 008ca5b492 feat: restore legacy fp8 handling when building with upstream ggml (#2001) 2026-09-21 00:48:30 +08:00
leejet 137f7409bb feat: add Qwen Image 2.1 support (#1994) 2026-09-20 22:51:21 +08:00
leejet 1330cebae8 feat: support building with upstream ggml (#1999) 2026-09-19 22:25:01 +08:00
leejet 17860c0e45 perf: parallelize host tensor elementwise and broadcast ops (#1998) 2026-09-19 21:46:16 +08:00
leejet 275ab58e01 perf: reduce CPU overhead in graph execution and sampling (#1997) 2026-09-19 18:32:58 +08:00
Fabrice Aneche d32b4e893b fix: prevent clip_preprocess center crop from exceeding the resized image (#1995) 2026-09-19 18:08:34 +08:00
leejet 3e037a81e4 perf: accelerate VAE direct 3D convolutions (#1996) 2026-09-19 17:50:01 +08:00
leejet adcac69650 perf: pad small attention heads to 64 for MMA Flash Attention (#1992) 2026-09-18 23:59:58 +08:00
leejet 656a1354c3 refactor: remove obsolete unused tensor filtering (#1984) 2026-09-18 23:41:37 +08:00
Lin Xuhao 269e726015 fix: honor flash attention flag in LLM text encoder attention (#1987) 2026-09-18 23:41:24 +08:00
leejet cc515a01f9 perf: eliminate temporary allocations in Philox rounds (#1982) 2026-09-17 02:09:12 +08:00
leejet 3161505fe8 fix: remove vision_model. from ununsed tensors (#1983) 2026-09-17 02:08:24 +08:00
leejet 59c23bce0d fix: use tokenizer-specific pre-tokenization rules (#1975) 2026-09-15 02:37:52 +08:00
Wagner Bruna 07a85c74cb feat: support Brownian tree noise in all noise injection samplers (#1899) 2026-09-15 02:37:31 +08:00
leejet f9ddc0f388 refactor: require external Gemma 2 and GPT-OSS tokenizers (#1974) 2026-09-15 01:25:13 +08:00
leejet 4964abdfc5 feat: support external Hugging Face tokenizer JSON files (#1973) 2026-09-15 00:27:39 +08:00
Санька Четвёртыйandleejet 42d6c0ab92 feat: Add generation parameters into video metadata (#1901)
Co-authored-by: leejet <leejet714@gmail.com>
2026-09-14 00:01:09 +08:00
leejet 5a5400bf0c fix: resolve MSVC narrowing conversion warnings (#1969) 2026-09-13 23:45:54 +08:00
fszontagh ca37fad89a fix: validate vision projector output dim against LLM hidden size (#1918) 2026-09-13 23:42:46 +08:00
Georgeandleejet 0bd72f075a feat: add Wan2.2 S2V (audio+img-to-video) support (#1925)
Co-authored-by: leejet <leejet714@gmail.com>
2026-09-13 23:33:50 +08:00
fszontagh 4a7da26b73 fix: bound plain-text runs in parse_prompt_attention regex (#1919) 2026-09-13 23:29:19 +08:00