Search:Reasoning

By relevance · Total 57 results Clear search

Nov 22026Monday

08:00
LMSYS BlogNov 2 08:00
LMSYS explains in detail how HiCache adapts to multi-parallelism and distributed inference
The LMSYS blog introduces how HiCache is implemented in context parallelism, pipeline parallelism, speculative decoding, unified KV layout, and P/D disaggregation scenarios, with the core being to keep cache decisions consistent across ranks. The iFlytek Spark MaaS team used HiCache plus external storage to serve GLM-5.1/5.2 CodingPlan workloads, raising the cache hit rate from about 30% to 95% and reducing median TTFT to under 3 seconds. In CP sharding tests, the GPU token hit rate increased by 40.48 percentage points, the host memory serving token volume dropped by 81.2%, and the average prefill queueing time decreased by 56.2%.
The LMSYS blog introduces how HiCache is implemented in context parallelism, pipeline parallelism, speculative decoding, unified KV layout, and P/D disaggregation scenarios, with the core being to keep cache decisions consistent across ranks. The iFlytek Spark MaaS team used HiCache plus external storage to serve GLM-5.1/5.2 CodingPlan workloads, raising the cache hit rate from about 30% to 95% and reducing median TTFT to under 3 seconds. In CP sharding tests, the GPU token hit rate increased by
Matched fieldsTitle
#company #Tutorials / practice #English text is being prepared.#Open-source ecosystem

Oct 292026Thursday

02:35
ARC PrizeOct 29 02:35
ARC Prize publishes Muse Spark Open reasoning scaling test results on ARC-AGI
ARC Prize published the test results of Meta's Muse Spark Open on the public ARC-AGI tasks: ARC-AGI-2 improves with reasoning scale, with medium averaging 55k tokens per task for 29.4%, high at 150k tokens for 51.0%, and xhigh at 180k tokens for 57.2%. ARC-AGI-1 already reached 92.8% at high and rose to 94.0% at xhigh. The model scores 57.5% on ARC-AGI-2 at $0.73 per task, and 89.0% on ARC-AGI-1 at $0.29 per task; its scores are close to those of Haiku 5.5 and GPT-6 Luna, but its cost is about 9 times theirs.
ARC Prize published the test results of Meta's Muse Spark Open on the public ARC-AGI tasks: ARC-AGI-2 improves with reasoning scale, with medium averaging 55k tokens per task for 29.4%, high at 150k tokens for 51.0%, and xhigh at 180k tokens for 57.2%. ARC-AGI-1 already reached 92.8% at high and rose to 94.0% at xhigh. The model scores 57.5% on ARC-AGI-2 at $0.73 per task, and 89.0% on ARC-AGI-1 at $0.29 per task; its scores are close to those of Haiku 5.5 and GPT-6 Luna, but its cost is about 9
Matched fieldsTitle
#company #Evaluations / benchmarks #Reasoning #Meta

Oct 242026Saturday

19:14
Ahead of AIOct 24 19:14
Sebastian Raschka releases an 11-hour video course on building a reasoning model from scratch
Sebastian Raschka has released 8 programming videos on building a reasoning model from scratch, totaling 11 hours, using Python and PyTorch to implement evaluation, inference-time scaling, reinforcement learning, and distillation starting from a pretrained Qwen3 0.6B model. The course first builds a math answer verifier and evaluates on MATH-500, then implements temperature sampling, top-p filtering, and self-consistency, followed by RLVR training with a simplified version of GRPO, and finally distills a small model using answers generated by DeepSeek R1. The content accompanies his book "Build a Reasoning Model (From Scratch)"; the early text generation and evaluation examples can run on a CPU, while later training experiments are recommended to use a GPU.
Sebastian Raschka has released 8 programming videos on building a reasoning model from scratch, totaling 11 hours, using Python and PyTorch to implement evaluation, inference-time scaling, reinforcement learning, and distillation starting from a pretrained Qwen3 0.6B model. The course first builds a math answer verifier and evaluates on MATH-500, then implements temperature sampling, top-p filtering, and self-consistency, followed by RLVR training with a simplified version of GRPO, and finally d
Matched fieldsTitle
#newsletter #Tutorials / practice #Reasoning #English text is being prepared.
07:19
Artificial AnalysisOct 24 07:19
Artificial Analysis measured Meta Muse Spark Open inference throughput
Artificial Analysis published preliminary test results on AA-AgentPerf for Meta's new open-weight model Muse Spark Open, an 816B-parameter, 42B-active MoE that scores 39 on the Artificial Analysis Intelligence Index. Tests show NVIDIA B300 and GB300 each reach about 150M tokens/s per megawatt under throughput-first configurations, with GB300 at about 108M and B300 at about 92M at 100 tok/s output speed; AMD MI355X reaches 70M tokens/s per megawatt at 65 tok/s. 46 of the model's 62 layers use 2,048-token sliding window attention, reducing KV cache requirements at 1M context by about 75% versus full attention; Meta released an NVFP4 checkpoint and AMD released an MXFP4 checkpoint.
Artificial Analysis published preliminary test results on AA-AgentPerf for Meta's new open-weight model Muse Spark Open, an 816B-parameter, 42B-active MoE that scores 39 on the Artificial Analysis Intelligence Index. Tests show NVIDIA B300 and GB300 each reach about 150M tokens/s per megawatt under throughput-first configurations, with GB300 at about 108M and B300 at about 92M at 100 tok/s output speed; AMD MI355X reaches 70M tokens/s per megawatt at 65 tok/s. 46 of the model's 62 layers use 2,0
Matched fieldsTitle
#editorial #Evaluations / benchmarks #Reasoning #English text is being prepared.

Oct 162026Friday

16:39
heise onlineOct 16 16:39
Aleph Alpha Releases Kolibri-1: A 78B-Parameter Open-Weight Model for German-Language Reasoning
Aleph Alpha has released the open-weight model Kolibri-1, with 78B total parameters. It uses a MoE architecture with 384 experts and 6 routed experts, activating only 3.46B parameters during inference, with a 256K Token context that can be extrapolated to about 1 million Tokens. The model focuses on German and English, and its tokenizer segments German words more token-efficiently; for example, the German Bundestag glossary requires only 1137 Tokens, whereas the compared DeepSeek-V4.1-Flash requires 1910 Tokens. Pretraining used 20 trillion Tokens, with data current as of June 18, 2026. The author's hands-on test found its answers correct and its German-language reasoning process evident, and stated that the Merlin-Arthur protocol makes the model hallucinate less in benchmarks, but its actual performance still needs to be verified in real-world use.
Aleph Alpha has released the open-weight model Kolibri-1, with 78B total parameters. It uses a MoE architecture with 384 experts and 6 routed experts, activating only 3.46B parameters during inference, with a 256K Token context that can be extrapolated to about 1 million Tokens. The model focuses on German and English, and its tokenizer segments German words more token-efficiently; for example, the German Bundestag glossary requires only 1137 Tokens, whereas the compared DeepSeek-V4.1-Flash requ
Matched fieldsTitle
#hot-signal #Model releases #Reasoning #English text is being prepared.
09:18
artificialanlysOct 16 09:18
English translation is being prepared.
Matched fieldsTitle
#editorial #Evaluations / benchmarks #Agents #English text is being prepared.
10:27
apple_machine_learning_researchOct 16 10:27Selected
English translation is being prepared.
Matched fieldsTitle
#lab #Tutorials / practice #English text is being prepared.#English text is being prepared.

Oct 132026Tuesday

20:33
xiaomi_mimo_blogOct 13 20:33
English translation is being prepared.
Matched fieldsTitle
#company #Tutorials / practice #English text is being prepared.#Reasoning

Oct 32026Saturday

05:09
basetenOct 3 05:09
English translation is being prepared.
Matched fieldsTitle
#a6-editorial #Tutorials / practice #Reasoning #English text is being prepared.
05:09
basetenOct 3 05:09
English translation is being prepared.
Matched fieldsTitle
#editorial #Tutorials / practice #Reasoning #English text is being prepared.

Oct 22026Friday

00:51
arenaOct 2 00:51
English translation is being prepared.
Matched fieldsTitle
#editorial #Evaluations / benchmarks #Agents #English text is being prepared.

Oct 12026Thursday

12:21
A6 a6-051 · Cerebras：BlogOct 1 12:21Selected
Cerebras explains model and reasoning-effort selection for GPT-5.6 Sol, Terra, and Luna
Cerebras published a post introducing the three GPT-5.6 models Sol, Terra, and Luna in the Codex subscription; the three are trained and deployed independently and come with reasoning-effort levels, used to trade off among speed, cost, and intelligence. On price, Terra is half of Sol, and Luna is one-fifth of Sol, for both short and long contexts. The article suggests starting most tasks with Luna and upgrading to Terra or Sol when progress stalls; Sol can reach 750 tokens/s on Cerebras, and it also explains how to use the five reasoning levels in Codex, from Light to Ultra.
Cerebras published a post introducing the three GPT-5.6 models Sol, Terra, and Luna in the Codex subscription; the three are trained and deployed independently and come with reasoning-effort levels, used to trade off among speed, cost, and intelligence. On price, Terra is half of Sol, and Luna is one-fifth of Sol, for both short and long contexts. The article suggests starting most tasks with Luna and upgrading to Terra or Sol when progress stalls; Sol can reach 750 tokens/s on Cerebras, and it
Matched fieldsTitle
#a6-editorial #Tutorials / practice #Reasoning #English text is being prepared.