GitHub Repo
Ollama 0.40.10 Adds Nemotron Bolt E4B Support and DFlash Speculative Decoding
Released on October 23, Ollama 0.40.10 adds NVIDIA Nemotron 3.5 Bolt E4B and its companion DFlash draft model. The update offers a new option for local agent models, but package size and input capability labels differ between Ollama and NVIDIA’s quantized weights, so verify them before deployment.
Ollama released version 0.40.10 on October 23. Its main update is support for NVIDIA Nemotron 3.5 Bolt E4B, released the same day, along with a DFlash draft model for speculative decoding. Developers can launch it with ollama run nemotron-3.5-bolt:e4b. Following the previous version’s addition of voice features, this is another update focused on running models locally. Release announcement
Bolt E4B is derived from Gemma 4 E4B through further post-training. It has about 8 billion total parameters and about 4 billion active parameters, and is aimed at use cases including multi-turn tool use, instruction following, and device assistants. NVIDIA’s model card says training used multi-environment GRPO reinforcement learning across tool use, conversations, and structured outputs. It will be worth seeing whether the model can sustain agent workflows and maintain output formats on limited hardware. NVIDIA model card
DFlash provides an additional draft model for speculative decoding: the draft model proposes candidate content, which the main model then verifies. Ollama’s model page lists it as a companion for faster generation and says E4B supports Windows, Linux, and Apple Silicon. However, the release announcement provides no comparable speed data for these platforms. Measure actual gains using your own prompt lengths, tool responses, and hardware. Ollama model page
Pay particular attention to the packages when deploying. Ollama’s e4b tag currently lists a size of about 7.2 to 7.9 GB, a 128K context window, and text/image input. NVIDIA’s “4GB” Q4_K_M model card, by contrast, lists text input and output, with up to 64K tokens each. These labels are not directly interchangeable, and the weight size should not be taken as the memory requirement for the entire service. Ollama package details, quantized model card
For Chinese-language agent applications, the next step is to test Traditional Chinese tool arguments, instruction retention across long conversations, and changes in latency and memory use when the draft model is enabled. Available sources confirm the model’s release and runtime support, but they are not enough to infer its quality or speedup on Chinese-language workloads.