Repository navigation
mlx-agent v0.6.0
-
Add
serve start --jit: the loopback endpoint stays reachable while its local model is unloaded; chat and completion requests load one owned worker from the confirmed local files, concurrent requests share the load, and active streams block unload. -
Add
serve unload(receipt-fenced,--expected-pid) to release JIT weights while the endpoint stays reachable, and separate residency inserve status. -
Add JIT memory policy:
--idle-timeout,--keep-loaded, and--min-headroom-gbonserve startandserve policy;serve policyupdates a running endpoint without a restart; cold loads check fresh OS headroom against estimated weights, runtime allowance, and reserve, and unknown measurements block loading. -
Fix:
convert describedecodes the answer with the tokenizer's special tokens skipped; moondream3 answers no longer start with<|md_reserved_4|>. -
Fix:
convert transcribewith Whisper vocabularies before large-v3 ends at the end of speech. mlx-audio resolved their missing<|nospeech|>to<|endoftext|>and suppressed it, so every transcript ran to the token limit with invented text; the runner uses<|nocaptions|>instead. -
Add
convert speak: synthesize one text with a converted text-to-speech model to a new WAV (voice, speed, language), returning audio length, synthesis time, real-time factor, and peak memory (read-only). -
Add
convert describe: answer one question about an image or a video with a converted vision-language model through its backend, returning token counts, throughput, and timing (read-only). -
Add
convert video: render one prompt with a converted text-to-video model to a new MP4 (size, frames, fps, steps, seed), returning frame count, duration, generation and load time, seconds per frame, peak memory, and pixel spread (read-only). -
Add the
mlx-videobackend (hash-lockedmlx-video-with-audio0.1.37) with at2vport:intake resolveandconvert starttake the original Wan2.1 T2V checkpoint (Wan-AI/Wan2.1-T2V-1.3B) to an MLX directory at 4 or 8 bits; new task type and backend categoryvideo_generation(pipeline tagtext-to-video). -
Add a DeepSeek-V4 DSpark drafter port (
deepseek_v4_dspark):convert start --ggufon llama.cpp'sdflashdrafter GGUF writes an MLX sidecar, tensors named as DeepSeek's checkpoint (mtp.Nasstages.N), FP8-sourced projections affine-quantized, MXFP4 routed experts repacked to MLXmxfp4exactly,config.jsonwith the DSpark fields andtext_config. dspark-DeepSeek-V4-Flash-0731-Q8_0 at 4 bits: 4,730 tensors, 10.67 GB; F32, BF16, and FP4 tensors equal DeepSeek-V4-Flash-0731's own safetensors. -
convert scanlabels speculative-decoding drafters (dflash,eagle3GGUFs; configs withdspark_target_layer_ids)speculative_draft(use casespeculative_decoding), reportsdraft(port,target,block_size), and keys a drafter apart from its target so it is never grouped as the target's duplicate. Drafters no port converts are refused at plan time (unsupported_draft). -
Committed provider packages hold one runtime per installed package instead of one per skill: tracked files under
providers/drop from 1871 to about 200.providers/claude/is removed (Claude Code and the installer'sclaudeprovider use the repository-root plugin files);providers/codexandproviders/agycarry one runtime at the package root, which their skills resolve as<plugin-root>/scripts/mlx-agent;providers/opencodeandproviders/agentskillscarry none. -
install agentskillscopies the root launcher and runtime into each installed skill, andinstall opencodecopies rootsrc/mlx_agent; thecp -R providers/agentskills/...route is removed from the docs. -
updateanduninstallremove receipt-owned files that the current manifest no longer declares (the per-skill runtime copies of earlier releases), refusing any whose content changed. Transaction changes acceptremove, receipted with a backup and restored by rollback. -
Scout command and skill text point at the reference packs under the package runtime root (
${CLAUDE_PLUGIN_ROOT}/src/...,<plugin-root>/src/...,<skill-dir>/src/...,<config-dir>/src/...) instead of a baresrc/.... -
Fix: the OpenCode
mlx_agent_commandtool accepts all seven capabilities (it accepted only scout, adopt, and wire while bench, doctor, watch, and fleet commands called it). -
Fix: the Codex plugin author is generated from the manifest publisher (
organization_url) instead of a hardcoded GitHub profile. -
Add the mflux backend (0.20.0, hash-locked) with an
image_generationcategory and a Qwen-Image 2.1 port (qwen_image_21): its converter merges an optional PEFT LoRA into the transformer (W + scale·B@A, rounded to the weight dtype), quantizes with mflux, saves mflux's component layout, and writes a rootconfig.json(components,quantization). Diffusers repos resolve bymodel_index.jsonclass (port_pipelines); curatedport_recipesname a repository's pinned base plus its LoRA (abenzerps/Qwen-Image-2.1-Uncensored-GGUF: Qwen/Qwen-Image-2.1@d26bb61 + its LoRA at 1.0, equal bit for bit to its BF16 GGUF on the checked tensors).intake fetchdownloads a recipe's base snapshot and LoRA;convert startreads both from the cache.intake resolvereportsrecipe. -
Add
convert generate: render one prompt with a converted image-generation model to a new PNG (size, steps, seed). -
convert scanlists outputs saved as per-component folders whenconfig.jsonnames thecomponents. -
Add the mlx-embeddings backend (0.1.0, hash-locked) with a
classificationcategory, and a Laya port (laya, convaiinnovations/laya root checkpoint): ModernBERT encoder plus the option-marker decision head, with the shipped temperatures. Repos without a rootconfig.jsonresolve through a port'sport_signatures;convert start --model-typeruns a port's own converter (port_convert);port_quantizetakes agroup_size(Laya: encoder and head linears at group 32). -
Subfolder checkpoints:
org/name/folderandtree/<rev>/<folder>links resolve against that folder's files (config, weights headers, port signature, download size);intake fetchdownloads only that folder;convert start --subfolderconverts from the cached folder with any backend's converter. Laya'smultilingual(mmBERT-base) andtyped-decisionscheckpoints convert this way. -
The convert receipt schema allows
backend(written by optional-backend conversions) andsource.subfolder; tests validate real receipts against it. -
Ports can declare the bit widths their conversions keep accurate (
port_bits);intake resolvereports them asq_bitsandconvert startrefuses others. Laya converts at 8 bits only: at 4 bits the multilingual checkpoint changes 3 of 25 reference answers. -
intake resolvereportsdownload_bytes, what a snapshot fetch takes;intake fetch --model-typedownloads only a ported type's files (Laya: 846 MB of the 2.4 GB repo). -
Add
convert decide: answer typed questions (choice, score, noul) about a state with a converted classification model through its backend (read-only). -
Task labels:
text-classificationandzero-shot-classificationrepos, and models a classification backend implements, areclassification(use cases: moderation, routing, classification). -
Add
convert transcribe: transcribe one audio file with a converted speech-to-text model through its backend (read-only). -
watch diffno longer reports a gated change when either side is unknown or missing. -
intake resolvereportsestimated_output_bytes(4- and 8-bit) for convertible repos from the safetensors headers; ports declare what they quantize (port_quantize). Edge0/Audio8-ASR-Infinite: 3.75 GB at 4 bits, matching the real conversion. -
Add backend model ports: mlx-agent ships an MLX implementation of Edge0/Audio8-ASR-Infinite (
audio8_asr_infinite: Voxtral Realtime audio tower, Qwen2 decoder with delay conditioning, projector, frame-length embedding, semantic VAD heads, streaming greedy decode with the 30 s rolling window) for the mlx-audio backend.intake resolvenow reports Audio8 as convertible through mlx-audio;convert start --backend mlx-audioinstalls the port into the backend before converting. Conversion quantizes the decoder only. -
Task labels: a model is vision-language only with a vision config or a vision module (from the mlx-vlm registry's vision files); GGUF main weights are never vision, and GGUF multimodal projectors (
clip/mmproj) are labelled other. -
Add local-path serving:
serve start --path DIRserves a local model
directory (e.g. a converted output outside the Hugging Face cache) through
the same preview → confirm → receipt flow as--repo. The two are mutually
exclusive; path plans are gated on the directory existing instead of cache
membership, receipts carrypathalongsiderepo, and status/stop argv
matching accepts either. -
Add
intake resolve: classify a pasted Hugging Face link and report which declared MLX backend (mlx-lm, mlx-vlm, mlx-audio) converts it, or which components have no MLX implementation, without downloading weights. -
convert scanitems carry atasklabel (type, use cases, source, confidence). -
Add
backend list|install|remove: optional mlx-vlm and mlx-audio converters in isolated, hash-locked environments (confirmation-gated; removal moves to the Trash). -
Add
convert start --backendfor repo conversions through an installed optional backend. -
Add
intake fetchandintake status: confirmation-gated, receipt-tracked downloads. -
intake fetchskips weight formats MLX converters never read (TensorFlow, Flax, ONNX, Core ML, GGUF) in snapshot downloads. -
Add
intake port-analysis: deterministic component, weight-prefix, and custom-code analysis for architectures no backend implements. -
Add
intake port-plan: draft a porting plan from that analysis with a locally served model (loopback endpoints only). -
Fix:
convert start --reponow passes--quantize; repo conversions were previously written unquantized.
mlx-agent-docs-v0.6.0.tar.gz
sha256 b98280508791e79d8a760b8a73c7def9522c31ff4490eb77b737be4154d8b0f9