Skip to content
#Open source / repos
October 19Monday
  1. Sakana AI Blog55

    Sakana AI proposes QD-Rec, a Quality-Diversity recommendation method that pairs candidates with LLM-written hypotheses

    Sakana AI developed QD-Rec, a recommendation method that treats a candidate plus a hypothesis explaining why to choose it as the unit of recommendation and uses Quality-Diversity search to accumulate high-quality, diverse proposals. In experiments on M&A, customer targeting and paper recommendation, QD-Rec built archives with higher average scores and larger size than a baseline that writes each hypothesis from scratch, and its code is released on GitHub. The paper reports cases where a hypothesis scored 1 out of 10 for one company was mutated and transferred to another candidate, raising the score to 5.

  2. inclusionAI Repositories58

    inclusionAI releases Ming-3.0-tts speech synthesis models in 0.5B dense and 7.9B MoE tiny variants

    inclusionAI released Ming-3.0-tts, an open-weight multilingual speech synthesis family with two models: Ming-3.0-tts-0.5b, a 0.5B dense model, and Ming-3.0-tts-tiny, a MoE model built on Ling V3 with 7.9B total and 1.3B active parameters. Both ship with inference code, a task cookbook and benchmark tables, and support voice cloning, text-based voice design, free-form natural-language delivery control, 14 inline paralinguistic tags, emotion and speed/pitch/volume control, plus Chinese, English, German, Spanish, French, Korean, Japanese, Sichuanese and Cantonese. On the October 2026 Artificial Analysis text-to-speech leaderboards, Ming-3.0-tts ranks 1st among open-weights models on Controlled Voice, 1st overall on Pronunciation Robustness at 93.2%, and 3rd among open-weights models on Provider Voice.

  3. inclusionAI Repositories58

    inclusionAI releases Ming-3.0-asr-tiny context-aware speech recognition model

    inclusionAI released Ming-3.0-asr-tiny, a speech recognition model that pairs the Ling-tiny-v3 language model with an audio encoder for Chinese and English transcription, dialect recognition and context ASR. Dialect evaluation covers 15 test sets including Hunan, Minnan, Cantonese, Chuanyu and Shanghai, and hotwords or preceding dialogue can be supplied in the prompt to resolve names and technical terms. The repository ships Transformers and SGLang serving guides, and code and weights are released under the MIT License.

October 15Thursday
  1. Mistral AI71

    Mistral releases Timestral time-series foundation models, topping GIFT-Eval, TIME and fev-bench

    Mistral released Timestral, its first family of time-series foundation models, in four sizes from 5M to 209M parameters. The largest model ranks first on all three main public benchmarks, GIFT-Eval, TIME and fev-bench, among pre-trained zero-shot models from a single checkpoint without ensembling, using two-thirds of the parameters of the previous state of the art. Timestral produces probabilistic multivariate forecasts with 21 quantiles, supports past-only and known-future covariates, runs on CPU without a GPU, and is available through Mistral Forge for fine-tuning, with a GitHub repository and technical report published under a modified MIT licence.

    Why it matters: Mistral's first time-series foundation model family claims top zero-shot rankings on GIFT-Eval, TIME and fev-bench with a 209M-parameter checkpoint, offering a size-accuracy tradeoff reference for forecasting work.

  2. inclusionAI Repositories67

    inclusionAI releases Ming-3.0-audio-flash and Ming-3.0-audio-tiny audio models

    inclusionAI released Ming-3.0-audio-flash, a 126B-parameter Mixture-of-Experts audio model with 6.5B activated parameters, alongside the compact Ming-3.0-audio-tiny at 9.3B parameters with 2.7B activated. The flash model covers Chinese and English speech recognition across dialects and noisy multi-speaker settings, long-audio understanding, non-speech sound and music perception, audio reasoning, and end-to-end spoken dialogue with optional synthesized speech output. It ships with vLLM-Omni, SGLang, and Transformers inference paths under the MIT License, with default Flash profiles using tensor parallelism across four 80 GB H800 GPUs at 128K context.

    Why it matters: The release details a 126B-parameter MoE audio model with 6.5B activated parameters, plus a 9.3B tiny variant, giving a concrete reference for open audio-language model sizing and deployment.

  3. Goodfire Research62

    Goodfire research essay argues neural networks are already clean and superposition was misread

    Goodfire researcher argues that the canonical evidence for polysemanticity and superposition in neural networks was misread, and that what looked like arbitrary packing is mostly a motif he calls recursive generalization. Retraining InceptionV1 with ordinary architectural improvements and no change to model size or training data made 99-100% of neurons in its last layer easily interpretable, and preliminary audits of Qwen3-32B estimate roughly 90% of neuron poles have one main job. The essay concludes networks use one learning principle rather than two, so interpretability can proceed by studying neurons and circuits directly instead of first training decomposition models such as sparse autoencoders.

    Why it matters: A first-hand reinterpretation of the field's canonical polysemanticity evidence, arguing that networks generalize recursively rather than pack unrelated concepts, which would make direct mechanistic study viable.

October 13Tuesday
  1. Amazon Science75

    Amazon Web Services researchers propose priming to convert pretrained Transformers into hybrid state space models without full retraining

    Amazon Web Services researchers introduce priming, a method that reuses pretrained Transformer weights to cut the pretraining cost of hybrid state space architectures by more than 99%, using under 0.5% of the source model's pretraining token budget. In controlled comparisons, Gated KalmaNet outperformed Gated DeltaNet and Mamba-2, and a 32B GKA-based hybrid matched its source Transformer on reasoning benchmarks while decoding up to twice as quickly at 128,000 tokens of context. The team released 8B and 32B primed hybrid models, training code, Triton kernels, and a vLLM serving plugin under the Apache 2.0 License.

    Why it matters: Amazon Web Services researchers show that reusing pretrained Transformer weights can cut hybrid architecture pretraining cost by over 99%, making controlled comparisons of state space layers practical.

September 29Tuesday
  1. OpenClaw Blog71

    OpenClaw Enterprise launches as an open source platform for managing persistent agents

    OpenClaw announced OpenClaw Enterprise (OCE), an open source, vendor neutral platform for managing persistent agents in sensitive environments, currently in development ahead of its 1.0 release later this year. OCE adds an enterprise control plane with multi-tenancy, hard security boundaries, sandboxing, LLM-based reviews and fine-grained permissions, and lets the harness, model and sandbox be swapped for third party or internal implementations. Organizations can self host it today via docker-compose for local development or kubernetes for internal deployment, and it is being piloted at companies including Red Hat and OpenAI.

    Why it matters: OpenClaw Enterprise is an open source control plane for persistent agents in sensitive environments, with multi-tenancy, hard security boundaries and auditability, now available for internal pilot workloads.

September 24Thursday
  1. Modal Blog58

    Modal and CMU's Full Stack Data Lab release Quail, an AI-SQL inference engine that jointly optimizes query planner and inference

    Modal and Carnegie Mellon University's Full Stack Data Lab built Quail, a query-aware inference engine for AI-SQL workloads that jointly optimizes the SQL query planner and the inference engine. On one multi-join query it processes over a billion tokens per minute per H100 GPU, more than 10x faster than Modal's vLLM baseline on the same hardware, at under 6¢ per billion tokens; on the team's new AI-SQL benchmark it runs 1.84x faster than vLLM geometrically averaged over tasks. Because AI-SQL filters and joins need only a single token per sequence, Quail skips decode, sampling and prefill-decode disaggregation, and it uses the known query plan to order requests and manage KV cache eviction.

September 23Wednesday
  1. Modal Blog58

    How Modal serves trillions of tokens for trillion-parameter coding agents

    Modal explains how it optimized Kimi K2.6 inference for coding agents, reaching 2.8x per-user and 5.6x cross-user performance per replica and running services at hundreds of billions of tokens per day. The work combines custom DFlash speculative decoding fine-tuned on coding traces, FP8 KV cache and NVFP4 shared-expert quantization, HiCache CPU-tier caching, and KV cache-aware load-based routing. Modal upstreamed its SGLang patches and released the configuration behind Modal Auto Endpoints.

September 1Tuesday
  1. Zyphra Research48

    Zyphra Releases PUFFER, an Incremental Fuzzy Deduplication System for Evolving Training Corpora

    Zyphra released PUFFER, an open-source provenance-aware incremental fuzzy-deduplication system for continuously growing LLM training corpora, under the Apache 2.0 license. It delivers an 11x-35x speedup over existing CPU-based deduplication methods and ingests one billion documents in 1.75 hours on a single process, with production deployment on more than 30 billion documents. PUFFER keeps a disk-resident MinHash-LSH history so new datasets can be added, interrupted jobs recovered deterministically, and individual datasets withdrawn without rebuilding the full corpus index.

July 29Wednesday
  1. Berkeley AI Research58

    From CUDA to MLX: K-Search Brings Kernel Optimization Expertise to Apple Silicon

    IBM Research extended K-Search, the evolutionary kernel search framework from UC Berkeley Sky Lab, with an MLX backend and a structured CUDA-to-MLX translation layer that adapts existing CUDA kernels into Apple Silicon kernels. On an MLX attention kernel the approach reached 0.97x the speed of Apple's native MLX Attention kernel, up from 0.26x without the translation context, and an evolved Mamba SSM kernel reached up to 20x prefill speedup over the community mlx-lm implementation on an M1 Max. The authors attribute the gain mainly to the quality of context and constraints given to the LLM rather than its ability to write Metal code, and say the method is not specific to MLX.

July 21Tuesday
  1. Modal Blog58

    Modal and Cognition launch Devin Outposts with modal-devin sandbox integration

    Modal is a launch partner for Devin Outposts, a Cognition feature that moves Devin's code execution into an environment the customer controls while Devin's reasoning stays in Cognition's cloud. The open-source modal-devin library and CLI runs each Outpost session in a Modal Sandbox, giving access to GPU-backed machines, custom images, workspace snapshotting and scale-to-zero, and it is live in alpha starting July 21, 2026. Setup uses uvx modal-devin init, which generates an orchestrator function and a worker session function that teams can keep customizing.

July 1Wednesday
  1. Modal Blog58

    Modal Research and HeavyBall's MRP brings multi-token prediction to diffusion LMs

    Modal Research and NYU Shanghai's HeavyBall Research introduce Multi-token Residual Prediction (MRP), a 3-layer module attached to a frozen diffusion LM backbone that predicts the inter-step logit residual instead of the full next-step distribution. On SDAR-4B GSM8K, MRP holds 70.9% accuracy at K=3 versus 5.9% for naive MTP distillation, and in SGLang it reaches up to 1.56x throughput losslessly in speculative mode or recovers up to +16 accuracy points under aggressive low-threshold dynamic decoding. The paper, code, SGLang implementation, and SDAR-MRP models on Hugging Face are released.

June 24Wednesday
  1. Modal Blog58

    Modal launches Auto Endpoints, citing speculative decoding for state-of-the-art inference latency

    Modal is launching Modal Auto Endpoints, which bring Modal's infrastructure scalability to state-of-the-art inference performance with a single click while keeping control of the code. Modal says the main latency win comes from speculative decoding with custom DFlash speculator models, which cut about 100ms of end-to-end latency in a Decagon Voice case study, roughly 40% of server-side decode latency, making it over 60ms faster than the fastest proprietary alternative. Modal also released open source DFlash speculator models with Z Lab and SGLang, including one for Qwen 3.5 397B-A17B that improves on MTP by over 50%.

June 1Monday
  1. Modal Blog46

    Modal: Reinforcement Learning for LLM Post-Training Is an Infrastructure Problem

    Modal says the bottleneck in RL post-training of LLMs is now infrastructure, not algorithms, and has released an open-source library drawn from running RL at scale. It details three pain points — glue code, cluster queueing and GPU under-utilization — and reports RDMA weight sync cutting Qwen3 8B update time from 2.62s to 41ms.

May 27Wednesday
  1. MiniMax Blog72

    MiniMax upgrades Agent as Mavis and launches Agent Teams for long-running tasks

    MiniMax upgraded MiniMax Agent under the new name Mavis and launched Agent Teams, which let multiple role-specific Agents run in parallel on the desktop for long, complex tasks. The system uses Leader, Worker and Verifier roles with a state-machine Team Engine that cycles tasks through producing, verifying and done, and it merges TokenPlan and Agent Plan into one subscription spanning CLI, API and Agent across M2.7, music, video and voice. MiniMax says the Agent will be open-sourced soon, expected alongside the MiniMax M3 model, while the desktop app is already available.

    Why it matters: MiniMax explains the design tradeoffs behind its Leader-Worker-Verifier Agent Team, including handoff, sharing and aggregation costs, offering a concrete reference for building multi-Agent runtimes.

April 29Wednesday
  1. Factory58

    Factory benchmarks 13 models on code review, finds GPT-5.2 best and MiniMax M2.7 cheapest

    Factory built an open-source benchmark testing 13 models on 50 real pull requests from Sentry, Grafana, Keycloak, Discourse and Cal.com, with each model reviewing every PR at least three times at high reasoning effort and an LLM judge scoring findings against a human-curated golden set. GPT-5.2 led at 60.5% F1 for $1.25 per PR, ahead of Claude Opus 4.6 at 59.8% F1 and $3.11 per PR, while MiniMax M2.7 reached 45.6% F1 at $0.15 per PR and Kimi K2.5 hit 51.9% F1 at $0.41. Factory says cost explains only about 21% of quality variance, and it open-sourced the golden set, evaluation scripts and a Droid Action GitHub Action supporting all 13 models.

  2. Modal Blog33

    Building an RL Theorem-Proving Workflow on Modal

    AE Studio details how it trained LLMs to prove math theorems with reinforcement learning on Modal, comparing Evolution Strategies against GRPO while keeping the training loop fixed. The setup runs vLLM inference on GPUs, Lean compiler verification in isolated CPU sandboxes, and a lightweight orchestrator, using per-function images, .map() fan-out, and Volumes for base weights.

March 13Friday
  1. Berkeley AI Research38

    Berkeley AI Research Introduces SPEX and ProxySPEX for Identifying Interactions at Scale for LLMs

    Berkeley AI Research researchers describe SPEX and ProxySPEX, algorithms that identify influential feature, data, and model-component interactions in LLMs at scales far beyond prior methods. SPEX reframes interaction search as sparse recovery using sparsity and low-degreeness, while ProxySPEX exploits hierarchy to match SPEX with about 10x fewer ablations. On sentiment analysis, SPEX matches Faith-Shap and Faith-Banzhaf faithfulness on short inputs and retains it as context scales to thousands of features.

March 4Wednesday
  1. Modal Blog22

    Modal Product Updates: Directory Snapshots, GLM-5 Endpoint and Billing API

    Modal added Directory Snapshots for Sandboxes, letting users snapshot a single directory and mount it into a new Sandbox independently of the base image. Modal also released a free public GLM-5 endpoint with Z.ai through the end of April, compatible with OpenCode, Claude Code, OpenClaw and the Vercel AI SDK. The modal billing report CLI and modal.billing.workspace_billing_report API are now generally available for Team and Enterprise workspaces.

February 24Tuesday
  1. Vapi Blog38

    Give Your OpenClaw Agent a Voice: Adding Phone Calls with Vapi Skills

    OpenClaw agents can now make and receive phone calls through Vapi's AgentSkills-compatible skills, installed in about five minutes via the npx skills CLI or by asking the agent directly. The Vapi pipeline targets sub-500ms voice responses and supports mid-call tool calling, with eight skills covering assistants, calls, phone numbers, tools, squads, webhooks and workflows.

November 13, 2025Thursday
  1. Modal Blog40

    How Decagon Shipped Real-Time Voice AI on Modal

    Decagon and Modal partnered to build Decagon Voice 2.0, cutting latency by 65% while improving intent recognition and response quality. Decagon fine-tuned compact open-source models with SFT and RL to surpass larger frontier models, while Modal trained a custom EAGLE3 draft model with 38% higher accept lengths and re-engineered SGLang with an async speculative decoding scheduler, boosting throughput up to 12%.

October 31, 2025Friday
  1. Modal Blog22

    Modal updates Volumes v2, JavaScript and Go SDKs to beta

    Modal has released Volumes v2 in open beta for all apps, with a filesystem rebuilt for higher throughput, faster random access, and true high-concurrency writes from hundreds of containers, plus no cap on stored files. Its JavaScript/TypeScript and Go SDKs also entered beta at v0.5, adding a unified Client object and support for Sandboxes, Functions, Images, and Volumes.

September 26, 2025Friday
  1. Modal Blog60

    Modal engineers reverse-engineer how Flash Attention 4 works

    Modal engineers read the released Flash Attention 4 source code and reverse-engineered how the Blackwell-optimized attention kernel works, since no official technical report exists yet. The write-up traces a tile through the kernel's warp-specialized asynchronous pipeline and details two math tricks: a cubic-polynomial software exponential that can replace SFU-based exponentiation, and a smarter online softmax rescaling rule that Tri Dao said cuts corrections by a factor of 10. FA4 is reported to be roughly 20% faster than Nvidia's closed-source cudnn attention kernels, and a backwards version was not yet available at the time of writing.

    Why it matters: Modal engineers read the released Flash Attention 4 source and explain its warp-specialized pipeline and two math tricks, offering a transferable view of where CUDA kernel design is heading.

September 22, 2025Monday
  1. Modal Blog38

    Modal Vibe: An AI Coding Platform Built on Modal Sandboxes That Scales to 1,000 Apps in 5 Minutes

    Modal built Modal Vibe, an open-source demo vibe coding platform on Modal Sandboxes that scales from 0 to 1,000 AI-coded apps in about five minutes. A load test creating 1,000 apps, each in its own Sandbox, took roughly 300 seconds, with the AI API as the rate-limiting step. Modal says the Team Plan can run 1,000 concurrent Sandboxes, enough for over one million sessions per month.

September 16, 2025Tuesday
  1. Modal Blog47

    Inside Modal Notebooks: How Modal Built a Cloud GPU Notebook That Boots in Seconds

    Modal detailed the engineering behind Modal Notebooks, its cloud Jupyter notebook that boots GPUs and custom images in seconds with real-time collaboration. Kernels run inside Modal Sandboxes, with a modal-kernelshim daemon translating Jupyter ZeroMQ messages into HTTP calls tunneled through Modal's control plane. A lazy-loading Rust FUSE container filesystem fetches file contents on demand through a tiered cache, avoiding the image-unpacking latency that can approach a minute for an 8 GB Python/ML image.

August 11, 2025Monday
  1. Modal Blog38

    Modal Product Updates: GPU Memory Snapshots, Notebooks, Service Tokens, and More

    Modal added GPU memory snapshots in alpha, cutting cold boots for GPU workloads by 10x by skipping steps like Torch compilation, enabled via the enable_gpu_snapshot option. The update also brings Modal Notebooks in beta for running Python in the browser, Service Tokens for workspace managers, and browser-generated UIs for deployed Functions.

April 30, 2025Wednesday
  1. Modal Blog38

    Modal launches alpha SDKs for JavaScript and Go

    Modal released alpha versions of its official SDKs for JavaScript and Go, letting developers call deployed Modal Functions and create isolated sandboxes from Node, Next.js, or Go services without writing Python. The SDKs cannot deploy functions, which still require Python, and only support JSON-like or bytes data types for now. Modal says it can run 10,000 sandboxes in either language without issues.

March 17, 2025Monday
February 21, 2025Friday
  1. Modal Blog35

    Modal open sources GPU Glossary under CC BY 4.0

    Modal has open sourced its GPU Glossary, an interlinked compendium of short documents on GPU programming topics, on GitHub under a Creative Commons BY 4.0 License. The glossary was first released two months ago to fill a gap in cross-stack GPU reference material, and community requests to contribute prompted the move. The repo's GitHub Issues already list planned extensions, including two sample matmul kernels and a section on Thread Block Clusters.

February 14, 2025Friday
  1. Modal Blog22

    Modal product updates: Logs v2, live container profiling, and region selection on all plans

    Modal shipped a refreshed logs dashboard with time-range input filtering, live container profiling for debugging stuck containers or inputs, and Debug Shells that run commands like vim, ps, and strace inside running containers. Region selection is now available on all plans, priced with a multiplier on base prices, and the client adds OIDC authentication for cloud bucket mounts via the new oidc_auth_role_arn field.