LoRA on a 16 GB card: what fits and what does not
A 7B model is the honest ceiling for LoRA fine-tuning on a 16 GB consumer card, not because anything...
#Fine-Tuning
323 articles tagged with Fine-Tuning
アヴェにはJuli Clover Appleはないってバルトさんが言ってましたOLED iPad Mini Fast Charge Needs a 35W Adapter, Sold Separately https://www.macru...
アヴェにはJuli Clover Appleはないってバルトさんが言ってましたOLED iPad Mini Fast Charge Needs a 35W Adapter, Sold Separately https://www.macrumors.com/2026/10/26/oled-ipad-mini-fast-charge-35w-adapter/#...
Kite's 1.5B model: 83 tokens/s on a CPU alone (M1 Max, Q4 GGUF, llama.cpp). Merging the adapter took MLX from 74 to 161 ...
Kite's 1.5B model: 83 tokens/s on a CPU alone (M1 Max, Q4 GGUF, llama.cpp). Merging the adapter took MLX from 74 to 161 tokens/s, and the 0.99 GB Q4 file kept its quality: 75.0% pa...
OLED iPad Mini Fast Charge Needs a 35W Adapter, Sold SeparatelyApple's new iPad mini is fast-charge capable, but not wit...
OLED iPad Mini Fast Charge Needs a 35W Adapter, Sold SeparatelyApple's new iPad mini is fast-charge capable, but not with the adapter in the box. Apple's specifications say the dev...
Building a Deterministic Financial Agent with Dual-Persona LoRA
In casual chat, a 0.01% rounding discrepancy is trivial. In a multi-billion dollar M&A due...
The Sloth's Guide to Support Ticket Triage: Benchmarking Zero-Shot vs. Few-Shot vs. QLoRA on Ollama
🦥 The Dreaded Chore: Why Humans Shouldn't Triage Support Tickets October 20 was...
PRITHIVSAKTHIUR/flux-3-image-klein-loras-fast: flux-3-image-klein-loras-fast is an interactive image editing, text-to-image synthesis, and dynamic LoRA execution studio powered by the black-forest-labs/flux-3-image-klein Distilled architecture (ModularPipeline).
flux-3-image-klein-loras-fast is an interactive image editing, text-to-image synthesis, and dynamic LoRA execution studio powered by the black-forest-labs/flux-3-image-klein Distil...
UGREEN’s 45W world travel adapter is back to its Prime Day lowUGREEN’s 45W 6-port travel adapter is down to $27.99 on Am...
UGREEN’s 45W world travel adapter is back to its Prime Day lowUGREEN’s 45W 6-port travel adapter is down to $27.99 on Amazon, returning to its all-time low for your global trips.ht...
OpenOPC-Shadow-Adapter is 89 days old, 1.1k stars, +838 this week. A zero-modification human-in-the-loop adapter for Ope...
OpenOPC-Shadow-Adapter is 89 days old, 1.1k stars, +838 this week. A zero-modification human-in-the-loop adapter for OpenOPC's agentic DAG runtime — the combination explains the ac...
dk23lhl/termius-mcp: Your Termius sessions, ready for AI. Read terminals, run commands, and track native SFTP uploads through MCP. Community adapter for macOS Termius 9.38.1.
Your Termius sessions, ready for AI. Read terminals, run commands, and track native SFTP uploads through MCP. Community adapter for macOS Termius 9.38.1.
Ollama LoRA Adapters No Longer Supported: What to Do
Ollama 0.34.1 removed the Modelfile ADAPTER instruction. The exact errors, and two tested ways to run a LoRA fine-tune: a merged GGUF or llama.cpp --lora.
dsh-comfyui-t/dsh-comfyui: DSH 里的 ComfyUI 绘图台：底模/LoRA 多选、提示词词库与中文转标签、预设、角色卡与风格卡、LLLite、出图后放大、视角三档、种子控制、音频无关。
DSH 里的 ComfyUI 绘图台：底模/LoRA 多选、提示词词库与中文转标签、预设、角色卡与风格卡、LLLite、出图后放大、视角三档、种子控制、音频无关。
"📡 OpenOPC-Shadow-Adapter hit our signals today: 1.1k stars, only 88 days old, +838 this weekhttps://olud.ai/project/ahm...
"📡 OpenOPC-Shadow-Adapter hit our signals today: 1.1k stars, only 88 days old, +838 this weekhttps://olud.ai/project/ahmadhassan-bted-openopc-shadow-adapter.html#OpenSource #AI #Gi...
MCP in one picture: a travel adapter.5 AI apps × 10 tools = 50 custom connectors.With one open standard: 5 + 10 = 15.Bef...
MCP in one picture: a travel adapter.5 AI apps × 10 tools = 50 custom connectors.With one open standard: 5 + 10 = 15.Before you plug in: trusted servers only, least privilege, prun...
Researchers propose PARADE, a composable parametric demographic injection framework using reusable LoRA modules to simul...
Researchers propose PARADE, a composable parametric demographic injection framework using reusable LoRA modules to simulate survey responses.Source: arXiv cs.CLhttps://arxiv.org/ab...
📡 Detected by our signals today: OpenOPC-Shadow-Adapter⭐ 1.1k stars · only 87 days old · +838 this week · accelerating« ...
📡 Detected by our signals today: OpenOPC-Shadow-Adapter⭐ 1.1k stars · only 87 days old · +838 this week · accelerating« Zero-modification Human-in-the-Loop adapter for OpenOPC's ag...
Your Go Service Gets a Third Front Door: An MCP Adapter That Never Touches the Domain
Add an MCP server to a Go service with the official go-sdk: one new adapter, no production changes to the domain, and a test that proves it.
Public Jeb answers general questions. Judge Jeb is an adapter on the same base, trained on the decisions one engine actu...
Public Jeb answers general questions. Judge Jeb is an adapter on the same base, trained on the decisions one engine actually asks and on the outcomes people approved.Its first trai...
艦長はいつも「United Statesは仲間だ！」と言うんですが……Apple's Wall-Mounted HomeView Ships Without a Power Adapter https://www.macrumors.com...
艦長はいつも「United Statesは仲間だ！」と言うんですが……Apple's Wall-Mounted HomeView Ships Without a Power Adapter https://www.macrumors.com/2026/10/15/wall-mounted-homeview-ships-without-power-adapte...
My LoRA improved routing and made JSON less reliable
A local MLX LoRA improved synthetic routing accuracy while reducing strict JSON validity. What the frozen outputs, corrected failures and CPU scoring tutorial actually show.
Локально ≠ безопасно: как скачанные ИИ-модели и LoRA-адаптеры крадут пароли, ключи и деньгиЗаметки разработчика мультимо...
Локально ≠ безопасно: как скачанные ИИ-модели и LoRA-адаптеры крадут пароли, ключи и деньгиЗаметки разработчика мультимодальной ИИ-студии с распределённой сетью вычислений - о бэкд...
justsmile3/PromptGenerator: stable-diffusion android prompt-generator danbooru localdream npu on-device-ai stable-diffusion-1-5 prompt-engineering offline-first android-app webview text-to-image ai-art i18n single-file no-gradle qualcomm genshin-impact lora
stable-diffusion android prompt-generator danbooru localdream npu on-device-ai stable-diffusion-1-5 prompt-engineering offline-first android-app webview text-to-image ai-art i18n s...
QLoRA settings, VRAM, GGUF export and a working Ollama Modelfile for a fine-tuned gpt-oss-20b, including the Harmony tem...
QLoRA settings, VRAM, GGUF export and a working Ollama Modelfile for a fine-tuned gpt-oss-20b, including the Harmony template and stop token it needs.#llm #machinelearning #ollama ...
Fine-Tune an LLM on Your Mac with MLX: QLoRA in 16 Seconds (Real Run, No Cloud)
I fine-tuned a language model on a Mac. The training took about 16 seconds, used about 2.5 GB of...