[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8},"actionData","errors","root",{"_109":110,"_111":112},"routes/embed.x.$handle.$name.$",{"_9":10,"_11":12,"_13":14,"_15":16,"_17":18,"_19":20,"_21":-5},"handle","MadeiInGermany","name","Dental-Community","path","frontend_html.ts","project",{"_22":45,"_11":12,"_33":46,"_47":-5,"_48":42,"_49":50,"_51":50,"_52":53,"_54":55,"_56":57,"_58":59,"_60":32,"_61":-5,"_62":-5,"_63":36,"_64":65,"_66":67,"_68":69,"_70":71,"_72":-5,"_73":74,"_75":74,"_76":77,"_78":77,"_79":77,"_80":74,"_81":74,"_82":83,"_84":85,"_86":87,"_88":42},"file",{"_22":23,"_11":14,"_24":25,"_26":-5,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":38,"_39":40,"_41":42,"_43":-5,"_13":44},"permalink","https://www.val.town/x/MadeiInGermany/Dental-Community/code/frontend_html.ts","extension","id","3281360f-b4a6-4cce-81cc-c897731b5cdf","fileId","01a020b8-e4a0-743e-8867-a60c3ff7073c","parentId","branchId","d18db6a1-e7c5-4b05-a87f-4b49ab8219bc","branchVersion",1,"latest",true,"createdAt",["D",1787255383090],"version",0,"fileType","script","content","// frontend_html.ts — HTML page builder for Open-Source-Lab on val.town\nimport { SHARED_CSS, CATALOG, CAT_NAMES, BENCH_DATA } from \"./frontend_styles.ts\";\n\nfunction esc(s: string) {\n return s.replace(/&/g,'&').replace(//g,'>').replace(/\"/g,'"');\n}\n\nfunction badge(tag: string) {\n const map: Record = {\n open:'badge-open',new:'badge-new',exp:'badge-exp',hot:'badge-hot',local:'badge-local'\n };\n const labels: Record = {\n open:'Open',new:'New',exp:'Experimental',hot:'Popular',local:'Local'\n };\n if (!map[tag]) return '';\n return `${labels[tag]}`;\n}\n\nfunction modelCard(m: any) {\n const init = m.n.replace(/[^A-Za-z0-9]/g,'').slice(0,2).toUpperCase();\n return `

${init}

${['open','new','hot','local'].filter(() => Math.random()>.6).map(badge).join('')}

${esc(m.n)}

${esc(m.src)}

${esc(m.p)}Params

${esc(m.c)}Context

${esc(m.l)}License

`;\n}\n\nfunction sectionCards(cat: string, title: string, limit = 6) {\n const items = CATALOG.filter(m => m.cat === cat).slice(0, limit);\n return `

${title}

\n View all\n

${items.map(modelCard).join('')}

`;\n}\n\nfunction libCards() {\n return CATALOG.map(m => {\n const init = m.n.replace(/[^A-Za-z0-9]/g,'').slice(0,2).toUpperCase();\n return `

${esc(m.n)}

${esc(m.src)}

\n ${CAT_NAMES[m.cat]||m.cat}\n

${esc(m.p)}Params

${esc(m.c)}Context

${esc(m.l)}License

`;\n }).join('');\n}\n\nfunction benchRows() {\n return BENCH_DATA.map(r => {\n const typePill = r.cat==='ref'\n ? `Baseline`\n : (r.os ? `Open` : `Hosted`);\n const cell = (v: number) => `${v.toFixed(1)}

`;\n return `\n

${esc(r.n)}

${esc(r.o)}

\n ${typePill}\n ${cell(r.mmlu)}${cell(r.gpqa)}${cell(r.he)}${cell(r.math)}${cell(r.avg)}\n`;\n }).join('');\n}\n\nexport function buildPage(chatModels: Array<{id:string,name:string,org:string}>) {\n const modelOptions = chatModels.map(m =>\n ``\n ).join('');\n\n const poeLinks = [\n {n:'Qwen3.5-27B',d:'Alibaba\\'s dense hybrid reasoning model.',url:'https://poe.com/Qwen3.5-27B'},\n {n:'Hermes-3-405B',d:'Nous Research instruction-following flagship.',url:'https://poe.com/Hermes-3-405B'},\n {n:'DeepSeek-V3',d:'671B MoE with advanced mathematics reasoning.',url:'https://poe.com/DeepSeek-V3'},\n {n:'Phi-4',d:'Microsoft small-footprint reasoning model.',url:'https://poe.com/Phi-4-OSL'},\n {n:'Mistral-NeMo',d:'12B collaborative build, multilingual tasks.',url:'https://poe.com/Mistral-NeMo'},\n {n:'Qwen2.5-Coder-32B',d:'Baseline workhorse for open coding.',url:'https://poe.com/Qwen2.5-Coder-32B'},\n ];\n\n return `\n\n\n\n\nOpen-Source-Lab — Local Model Catalogue\n${SHARED_CSS}\n\n\n\n\n\n\n\n

Model Catalogue

A place
to test
local models

Curated local and open-source AI models. From community workhorses to experimental frontier architectures.

\n Browse 50 Models\n Try Live Chat\n

Test Local ModelsTry models instantly without installing them.

Check the SpecsCompare parameters, context sizes, find your niche.

Exchange KnowledgeDiscuss hardware with open source enthusiasts.

\n\n\n

Model of the Month

Spotlight — June 2026

GLM-5.1

\n Open\n New\n Popular\n

Zhipu AI's agentic coding flagship posts SOTA on SWE-bench Pro and sustains thousands of tool calls. MIT-licensed, fully self-hostable for end-to-end project execution, debugging, and log analysis at scale.

754BParams

205KContext

MITLicense

MoEArch

\n Test Model\n Try Live Chat\n

\n\n\n

\n \n \n \n \n \n \n \n \n

\n\n\n

\n Key:\n

OpenApache / MIT / similar

NewReleased last 90 days

ExperimentalEarly / research stage

PopularCommunity favourite

LocalConsumer hardware

\n\n${sectionCards('text','Language & Reasoning Models')}\n${sectionCards('code','Code & Engineering Models')}\n${sectionCards('vision','Vision & Multimodal',7)}\n${sectionCards('audio','Audio & Embedding Models')}\n${sectionCards('experimental','Experimental & Research')}\n\n\n

Ready to Go Local?

Start running open models today. No API keys, no subscriptions, no data leaving your machine.

\n Try Live Chat\n View Benchmarks\n

\n\n\n\n\n\n\n\n\n\n\n\n\n\n

\n \n

\n \n \n \n

1. Pick an inference engine Start here

Your engine determines speed, hardware support and ease of use. The most popular options in 2026:

Ollama — easiest entry point; one-line model pulls and a built-in API. Great for beginners and Mac users.
llama.cpp — C/C++ core powering most local tooling. Runs GGUF models on CPU, CUDA, Metal, Vulkan and ROCm.
vLLM — high-throughput GPU serving with PagedAttention; ideal for multi-user or production workloads.
LM Studio — polished desktop GUI for browsing, downloading and chatting with GGUF models, no terminal needed.

$ ollama run qwen3.5:27b\n$ ./llama-server -m model.gguf -ngl 99 -c 8192

2. Understand quantization

Quantization shrinks weights from 16-bit down to fewer bits, trading a little quality for big memory savings.

Q8_0 — near-lossless, ~1 byte/param. Use when you have the VRAM.
Q4_K_M — sweet spot. ~4.5 bits/param with minimal quality loss; the default most people run.
Q3 / Q2 — emergency tier for fitting big models on small cards; noticeable degradation.
Prefer a larger model at Q4 over a smaller at Q8 when quality matters.

3. Memory math (VRAM estimate)

At Q4 you need roughly 0.55 GB per billion parameters, plus context overhead.

7B @ Q4 → ~5 GB · fits an 8 GB card.
14B @ Q4 → ~9 GB · fits 12 GB.
32B @ Q4 → ~20 GB · needs 24 GB (RTX 4090 / 3090).
70B @ Q4 → ~42 GB · dual-GPU or 48 GB+, or offload to RAM.

Long context inflates the KV cache; a 128K window can add several GB on top.

4. Hardware tips

Apple Silicon shines for local LLMs thanks to unified memory — M-series with 64GB+ runs large MoE models comfortably.
NVIDIA remains fastest via CUDA; the 3090/4090's 24GB is the consumer sweet spot.
CPU + RAM offload works for big models but is slow; expect single-digit tokens/sec.
MoE models activate only a fraction of params per token, so they run far faster than their total size suggests.

5. Performance tips

Set -ngl / GPU layers as high as your VRAM allows to offload the whole model to GPU.
Lower temperature (0.1–0.3) for code and factual tasks; raise it (0.7–1.0) for creative writing.
Enable flash attention and KV-cache quantization to fit longer contexts.
For reasoning models, allow plenty of output tokens so chain-of-thought isn't cut off.

\n\n\n

\n \n

\n \n \n \n

\n \n \n \n \n \n

\n \n \n \n \n \n \n \n \n \n \n ${benchRows()}\n

Model	Type	MMLU	GPQA	HumanEval	MATH	Avg

MMLU = general knowledge · GPQA = graduate-level science (Diamond) · HumanEval = code generation · MATH = competition mathematics. Scores are representative composite figures from public reports — verify against primary sources before citing.

\n\n\n\n\n\n\n\n\n\n\n\n\n\n`;\n}","deleted",false,"label",[14],"0745959c-6279-46f7-ae8a-8d693b4d6bcb",["D",1787255383090],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","https://poe.com/Open-Source-Lab","imageUrl","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/ad613932-9c8c-47ad-961b-2ceeb5e3af00/square512?exp=1789541999&sig=4fdd0ab7c337a87c1f61bd649a33e2b6edb45d23d676809cb6049dc799a591ac","userId","01995b22-1b41-771e-a0ac-a4cfc92d3606","pinnedFileId","01a020b8-e4a0-743e-8867-ad0a2d2ef3a2","pinPreview","pinMigrated","screenshotVersion","likeCount","orgId","01995b22-1b41-771e-a0ac-a96f3508eae1","tursoName","0745959c627946f7ae8a8d693b4d6bcb","tursoHostname","0745959c627946f7ae8a8d693b4d6bcb-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789520505489],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_9":10,"_108":104},"org",{"_9":10,"_101":102,"_103":104,"_56":57,"_105":106,"_107":42},"branch",{"_35":30,"_11":89,"_22":28,"_90":45,"_91":92,"_93":30,"_94":50,"_95":96,"_97":98,"_99":12,"_100":10},"isOwned","main","projectId","forkedBranchId","c978b0db-3948-478a-b669-5d839bc00fa2","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","b812ec28-5b0b-46fd-9657-66c09ad5be0a","forkedProjectOrgId","e54478ee-7c91-4914-a560-525ba174a661","projectName","userHandle","displayName","Made in Germany","avatar","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/e077fc84-719f-4dd3-3403-3eb925812100/square512?exp=1789541999&sig=97b2adaf5af7fb6391e30784416bd1d5e6abced802d673236166f7d63d2a0296","tier","free","canRestrictVals","profileImageUrl","ctx",{"_82":-5,"_142":143,"_144":145,"_146":147,"_148":42,"_149":42,"_150":42,"_151":32,"_152":-5,"_153":-7,"_154":-7,"_155":36,"_156":42,"_157":42,"_158":-5,"_159":42,"_160":42,"_161":42,"_162":-5},"ENV",{"_113":114,"_115":116,"_117":89,"_118":119,"_120":121,"_122":123,"_124":125,"_126":127,"_128":129,"_130":131,"_132":133,"_134":135,"_136":137,"_138":139,"_140":141},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","RENDER_GIT_COMMIT","e87eb04e707cf64e7e532c8e8ec395f209e17f15","RENDER_GIT_BRANCH","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","CLOUDFLARE_TURNSTILE_SITE_KEY","0x4AAAAAACtaIZJH7s9rBMZ6","GOOGLE_CLIENT_ID","875989696899-jtsqkhlbdv5jrlsk8631p7ipjg3c6o62.apps.googleusercontent.com","STRIPE_PUBLISHABLE_KEY","pk_live_51N2zvWEpZTTLipIkVEwXrUG9AT0yeOOEYxszarZT1jpsZV58Ly6wlg3SltZpHN26CUalOBmUSY7adQRQiSkCZz2500qSXNkf8Y","API_URL","https://api.val.town","orgs",[],"preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","preferencesCloseBrackets","preferencesChatPanelState","preferencesDprint","preferencesDenofmt","notificationCount","flagDenols","isBusinessOrgMember","businessOrgHandle","isBusinessOrgAdmin","isHigherTownieTier","navCollapsed","currentOrgId"]

${title}

A placeto testlocal models

Model of the Month

Ready to Go Local?

1. Pick an inference engine Start here

2. Understand quantization

3. Memory math (VRAM estimate)

4. Hardware tips

5. Performance tips

A place
to test
local models