←Search results

Search “DeepSeek”

By relevance · Total 1346 results Clear search

Thursday

  3. A6 a6-051 · Cerebras：Blog

    Cerebras: Fast Inference Finds Its Rhythm

    Cerebras looks back at 2025: DeepSeek R1 Llama 70B runs at over 1500 tokens/second on its chips, about 57 times that of GPUs; independent testing by Artificial Analysis shows its inference performance exceeds NVIDIA Blackwell's record. Cognition's SWE-1.5 and SWE-grep series achieve frontier-level coding performance at up to 13 times the speed of general-purpose models. Cerebras has launched a 40+ exaflop data center in Oklahoma and is deploying more than 7 data centers across North America and Europe.

    Cerebras looks back at 2025: DeepSeek R1 Llama 70B runs at over 1500 tokens/second on its chips, about 57 times that of GPUs; independent testing by Artificial Analysis shows its inference performance exceeds NVIDIA Blackwell's record. Cognition's SWE-1.5 and SWE-grep series achieve frontier-level coding performance at up to 13 times the speed of general-purpose models. Cerebras has launched a 40+ exaflop data center in Oklahoma and is deploying more than 7 data centers across North America and

    Matched fieldsSummary、Body

    #a6-editorial#Industry news#Reasoning#English text is being prepared.
  4. A6 a6-047 · Baseten Engineering Blog

    Baseten Engineering Blog: A Roundup of Model Performance Optimization Practices

    The model performance section of the Baseten engineering blog rounds up several inference optimization efforts, including speeding up pyannote's speaker diarization model by 9.6x, speeding up Kimi K3 tokenization by 18x under million-token agent workloads, and performing inference engineering optimization for DeepSeek V4 Pro 0813. The section also covers delta weight synchronization optimization for hosted RL rollouts, Coval voice AI benchmarking, and the practice of building APIs for Kimi K3 and GLM-5.2.

    performance section of the Baseten engineering blog rounds up several inference optimization efforts, including speeding up pyannote's speaker diarization model by 9.6x, speeding up Kimi K3 tokenization by 18x under million-token agent workloads, and performing inference engineering optimization for DeepSeek V4 Pro 0813. The section also covers delta weight synchronization optimization for hosted RL rollouts, Coval voice AI benchmarking, and the practice of building APIs for Kimi K3 and GLM-5.2.

    Matched fieldsSummary、Body

    #a6-editorial#Industry news#English text is being prepared.
  5. A6 a6-047 · Baseten Engineering Blog

    Baseten Engineering Blog AI Models Article List

    The AI Models section of the Baseten Engineering Blog brings together multiple articles on models and inference engineering, including NVIDIA Nemotron 3 Diarization, which achieves real-time speaker labeling at one cent per audio hour; DeepSeek-V4.1-Flash, which offers more efficient prefill for coding agents; and introductions to new models such as NVIDIA Nemotron 3.5 Lightning, Nemotron 3.5 ASR Streaming, and Nemotron 3 Embed. It also covers topics such as GLM-5.3 post-training scaling, GLM 5.2 vision capabilities, Laguna S 2.1, and recommendations for open-source post-trained models.

    ings together multiple articles on models and inference engineering, including NVIDIA Nemotron 3 Diarization, which achieves real-time speaker labeling at one cent per audio hour; DeepSeek-V4.1-Flash, which offers more efficient prefill for coding agents; and introductions to new models such as NVIDIA Nemotron 3.5 Lightning, Nemotron 3.5 ASR Streaming, and Nemotron 3 Embed. It also covers topics such as GLM-5.3 post-training scaling, GLM 5.2 vision capabilities, Laguna S 2.1, and recommendations

    Matched fieldsSummary、Body

    #a6-editorial#English text is being prepared.

Tuesday

Saturday

  1. Huxiu

    After Nvidia chips became hard to buy, what keeps China's large models running?

    A Huxiu analysis points out that domestic computing power can now cover large model operation: Zhipu's domestic-chip inference cluster has reached a scale of 100,000 cards, and Meituan's LongCat-2.0 has also completed training on domestic computing power. However, Ascend 910C achieves an MFU of about 30%-35% in trillion-parameter post-training, lower than H100's 40%-50%. The article breaks computing efficiency into three variables—chips, software, and models—and compares the four paths of Zhipu, MiniMax, DeepSeek, and Kimi, concluding that Zhipu has the highest short-term certainty, while DeepSeek has the greatest medium-term elasticity. The author proposes three tracking indicators—capacity, efficiency, and cost—and argues that only when all three improve simultaneously can domestic computing power shift from a backup option to a cost advantage.

    r post-training, lower than H100's 40%-50%. The article breaks computing efficiency into three variables—chips, software, and models—and compares the four paths of Zhipu, MiniMax, DeepSeek, and Kimi, concluding that Zhipu has the highest short-term certainty, while DeepSeek has the greatest medium-term elasticity. The author proposes three tracking indicators—capacity, efficiency, and cost—and argues that only when all three improve simultaneously can domestic computing power shift from a backup

    Matched fieldsSummary、Body

    #hot-signal#Industry news#English text is being prepared.#English text is being prepared.

Friday

  2. BBC Chinese

    US-China AI Race: Is the World Becoming More Dangerous or Safer?

    A top AI safety researcher at Anthropic warns that AI technology is advancing at an alarming rate and that within the next decade there is a probability as high as 10% that it "could lead to the extinction of all humanity." This has sparked a debate over whether the United States should slow down AI development and set regulatory "guardrails"; Trump believes that slowing the pace could cede dominance to competitors. Chinese AI services such as DeepSeek and Moonshot AI show that Chinese competitors' capabilities may already be comparable to those of leading US AI labs.

    ing at an alarming rate and that within the next decade there is a probability as high as 10% that it "could lead to the extinction of all humanity." This has sparked a debate over whether the United States should slow down AI development and set regulatory "guardrails"; Trump believes that slowing the pace could cede dominance to competitors. Chinese AI services such as DeepSeek and Moonshot AI show that Chinese competitors' capabilities may already be comparable to those of leading US AI labs.

    Matched fieldsSummary、Body

    #chinese#Industry news#Safety / alignment#OpenAI

Wednesday

Tuesday

Monday

Sunday

Friday