Bonsai 2 27B: 262k context (trained max) on 12 GB, 54 to 97 tok/s decode, 2x prefill, q8_0 KV recipe, grammar tool calls 9/9. Same ternary weights. Windows bundle + receipts.
-
Updated
Sep 30, 2026 - Python
#
rtx-4070
Here are 6 public repositories matching this topic...
Prebuilt CUDA wheels (sm_89) + build guide to run Microsoft TRELLIS.2 and TencentARC Pixal3D image-to-3D in ComfyUI on NVIDIA RTX 40-series (Ada) under Debian 13 / Linux. Fixes the 'no kernel image is available' (sm_120 vs sm_89) problem.
debian cuda pytorch nvidia 3d-generation ada-lovelace image-to-3d comfyui prebuilt-wheels rtx-4090 trellis2 rtx-4070 pixal3d sm-89
-
Updated
Jul 26, 2026 - Shell
Fork of karpathy/nanochat pushed onto one RTX 4070 12GB: a 910M-param base model from 530 GPU-hours, OOM-safe training automation, and 14 documented negative results.
-
Updated
Sep 9, 2026 - Python
Reproducible Qwen3.5 MTP and D-Flash benchmarks on an RTX 4070: throughput, acceptance rate, quantization, and language effects.
-
Updated
Sep 10, 2026 - Shell
Measured llama.cpp and quantization results on Tesla V100 (sm_70), Pascal GTX 1070, and RTX 4070 - hardware most projects don't test on
pascal nvidia quantization volta v100 mixture-of-experts gpu-benchmark humaneval llama-cpp local-llm llm-inference gguf speculative-decoding tesla-v100 coding-agent swe-bench diffusion-language-model sm70 rtx-4070 gtx-1070
-
Updated
Oct 1, 2026 - Python
Локальный LLM-провайдер для проекта slovo — Ollama c Laguna XS.2 (33B/3B MoE) в Docker с проброшенной NVIDIA GPU. Оптимизирован под single-user агентские сессии на RTX 4070 Ti SUPER + i9-11900K.
docker-compose moe code-generation nvidia-gpu mixture-of-experts wsl2 laguna llm local-inference ollama agentic-coding rtx-4070
-
Updated
Aug 2, 2026 - Python
Add this topic to your repo
To associate your repository with the rtx-4070 topic, visit your repo's landing page and select "manage topics."