[{"data":1,"prerenderedAt":77},["ShallowReactive",2],{"list-learn":3},[4,16,26,35,44,52,61,69],{"collection":5,"slug":6,"path":7,"title":8,"seoTitle":9,"description":10,"summary":11,"primaryKeyword":12,"cluster":13,"published":14,"updated":14,"reviewed":14,"readingMinutes":15},"learn","gguf-quantization","\u002Flearn\u002Fgguf-quantization\u002F","GGUF quantization explained","GGUF Quantization Explained: Q4_K_M, Q8_0 and File Size","What GGUF is, what Q8_0, Q6_K, Q5_K_M, Q4_K_M and IQ4_XS mean, and how to estimate file size from bits per weight, checked against real Hugging Face files.","GGUF is the single-file model format of the ggml project, used by llama.cpp. A quantization name such as Q4_K_M says how many bits each weight gets and how they are grouped. File size is roughly parameters × bits per weight ÷ 8, so a 27B model at Q4_K_M, which averages a little under 5 bits per weight, comes to about 16.5 to 17 GB. Real files differ because tensors are quantized differently.","gguf quantization explained","deployment","2026-11-14",9,{"collection":5,"slug":17,"path":18,"title":19,"seoTitle":20,"description":21,"summary":22,"primaryKeyword":23,"cluster":24,"published":14,"updated":14,"reviewed":14,"readingMinutes":25},"evaluate-llm-for-coding","\u002Flearn\u002Fevaluate-llm-for-coding\u002F","How to evaluate an LLM for coding","How to Evaluate an LLM for Coding on Your Own Tasks","A practical method to evaluate an LLM for coding: real tasks, unit tests, pass@k, contamination checks, cost per solved task, latency, tool calls and run logs.","To evaluate an LLM for coding, test it on tasks drawn from your own work and score each answer by running unit tests, not by reading it. Sample each task several times to estimate pass@1 and pass@k, check for contamination, and record cost per solved task, latency and tool-calling reliability. Log every setting so the run can be repeated, and treat model-card and leaderboard claims as claims until you have reproduced them.","evaluate llm for coding","evaluation",8,{"collection":5,"slug":27,"path":28,"title":29,"seoTitle":30,"description":31,"summary":32,"primaryKeyword":33,"cluster":13,"published":14,"updated":14,"reviewed":14,"readingMinutes":34},"local-vs-cloud-llm","\u002Flearn\u002Flocal-vs-cloud-llm\u002F","Local vs hosted uncensored LLMs","Local vs Hosted Uncensored LLM: Privacy, Cost, Upkeep","Run a refusal-reduced model yourself or call a hosted API? Privacy, latency, upkeep and a worked cost comparison in which every input is a stated assumption.","Run the model locally when prompts must stay on hardware you control, your volume is high and steady, or you need one exact variant file, and you can absorb the upkeep. Use a hosted API when volume is low or irregular, the model does not fit your hardware, or your time costs more than the tokens. A hosted API is only as private as its operator's policy, so read it.","local vs hosted uncensored llm",7,{"collection":5,"slug":36,"path":37,"title":38,"seoTitle":39,"description":40,"summary":41,"primaryKeyword":42,"cluster":43,"published":14,"updated":14,"reviewed":14,"readingMinutes":34},"llm-model-licenses","\u002Flearn\u002Fllm-model-licenses\u002F","Open-weight model licenses compared","Open-Weight Model License Comparison: MIT, Apache, Llama","Commercial use, notices, naming rules, user thresholds and use policies under MIT, Apache 2.0, Llama 3.3, Llama 3.1 and Gemma terms, and what variants inherit.","GLM models use MIT; Gemma 4, Qwen3.8 and gpt-oss-20b use Apache 2.0. Both let you use, modify and sell the weights if you meet their notice conditions. The Llama 3.3 and 3.1 licenses add attribution, a naming rule, a 700-million-user threshold and an acceptable use policy. Gemma 3's terms add a prohibited use policy that must travel with every copy. A refusal-reduced variant carries its base model's terms. Not legal advice.","open weight model license comparison","licensing",{"collection":5,"slug":45,"path":46,"title":47,"seoTitle":48,"description":49,"summary":50,"primaryKeyword":51,"cluster":43,"published":14,"updated":14,"reviewed":14,"readingMinutes":34},"open-weight-vs-open-source","\u002Flearn\u002Fopen-weight-vs-open-source\u002F","Open-weight vs open-source AI","Open-Weight vs Open-Source AI: What the License Allows","Open-weight means you can download the weights. Open source AI means meeting the OSI definition. What MIT, Apache 2.0, Llama and Gemma terms actually allow.","Open-weight means the trained parameters can be downloaded. Open source AI is a stricter claim: under the OSI Open Source AI Definition, the terms must let anyone use, study, modify and share the system, and the release must include data information, training code and the parameters. A model under MIT or Apache 2.0 can still fall short, and community licenses such as Llama's add use restrictions the definition does not allow.","open weight vs open source ai",{"collection":5,"slug":53,"path":54,"title":55,"seoTitle":56,"description":57,"summary":58,"primaryKeyword":59,"cluster":60,"published":14,"updated":14,"reviewed":14,"readingMinutes":34},"uncensored-vs-abliterated","\u002Flearn\u002Funcensored-vs-abliterated\u002F","Uncensored vs abliterated LLMs","Uncensored vs Abliterated LLM: How They Differ","Abliterated names one weight-editing method; uncensored is a looser label. Compare data, compute, side effects and checks, and learn to read a model card.","\"Abliterated\" names a method: a model's weights were edited to suppress the internal direction linked to refusals. \"Uncensored\" names a goal, and a model sold under it may have been abliterated, fine-tuned on new data, processed with an automated tool such as Heretic, or changed in a way the creator does not describe. The method, not the name, tells you what to expect and what to test.","uncensored vs abliterated llm","terminology",{"collection":5,"slug":62,"path":63,"title":64,"seoTitle":65,"description":66,"summary":67,"primaryKeyword":68,"cluster":60,"published":14,"updated":14,"reviewed":14,"readingMinutes":25},"what-is-abliteration","\u002Flearn\u002Fwhat-is-abliteration\u002F","What is abliteration?","What Is Abliteration? Refusal Directions in LLMs","Abliteration edits an LLM's weights to suppress the activation direction tied to refusals. How it works conceptually, how it is measured and its limits.","Abliteration is a way to make an open-weight language model decline fewer requests by editing its weights instead of retraining it. It rests on research showing that refusal in many chat models is tied to one direction in the model's internal activations. That direction is estimated by contrasting two sets of prompts and then removed from the weights. The edit is cheap, but it is approximate, model-specific and can reduce capability.","what is abliteration",{"collection":5,"slug":70,"path":71,"title":72,"seoTitle":73,"description":74,"summary":75,"primaryKeyword":76,"cluster":60,"published":14,"updated":14,"reviewed":14,"readingMinutes":34},"what-is-uncensored-ai","\u002Flearn\u002Fwhat-is-uncensored-ai\u002F","What is uncensored AI?","What Is Uncensored AI? Meaning, Types and Limits","Uncensored AI usually means an open-weight model changed to decline fewer requests. The main types, what the label does not promise and who stays responsible.","\"Uncensored AI\" is a community label, not a technical standard. It usually means an open-weight language model that someone other than its original developer has changed so that it declines fewer requests, through fine-tuning, abliteration, an automated tool such as Heretic, or an undisclosed method. The label does not mean the model has no rules, is more capable or keeps your data private, and the base license and the law still apply.","what is uncensored ai",1794622337619]