Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › Llama-2 family

Llama-2-70b-hf

Hugging Face ↗ Follow changes

meta-llama Llama 2 Community License out 3 years ago

69.0B parameters, text in, text out.

Verified 2026-10-25 Rank #389 in LLMs · score 79.8 method v1 Sources 3 Downloads 30d 22.3k · #448 of 619 tracked

Can I run it?41GB at INT4Estimatefits a 64 GB Mac (unified memory)method
Can I use it?Llama 2 Community LicenseAttribution; >700M MAU clause; acceptable-use policylicence file2026-10-24
Do people use it?22.3kdownloads, 30 days#448 of 619 we track · below the median+0 likes, downloads +12% this week856 likesHugging Face2026-10-24

Worth reading

Picked by rule from 29 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 955 HN pts

Fine-tune your own Llama 2 to replace GPT-3.5/4

Fine-tuning open-source Llama 2 models can replace GPT-3.5 and GPT-4 for specific tasks, with a fine-tuned Llama 7B matching GPT-4 labels 95% of the time on a test set.

news.ycombinator.com 2023-09-12 955 HN pts · 181 comments 10k words about the Llama 2 family family
Latest · 3 years ago

Run Llama 2 uncensored locally

Ollama enables local execution of uncensored Llama 2 variants, such as the 7B model fine-tuned using Wizard-Vicuna conversation dataset and Nous Hermes Llama 2 13B fine-tuned on over 300,000 instructions.

  • The uncensored model provides a specific recipe for dangerously spicy mayo including ingredients like cayenne pepper and paprika.
  • The uncensored model describes the manufacturing process for Tylenol by combining paracetamol with aspirin, caffeine, and diphenhydramine.
ollama.ai 2023-08-02 722 HN pts · 212 comments 1.3k words about the Llama 2 family family

25 best of 29 tracked

The 2 picks above, plus these 23. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. ArticleLlama 2ai.meta.com · 2023-07-18 · 2,268 HN pts · official
  2. CodeLlama2.c: Inference llama 2 in one file of pure Cgithub.com · 2023-07-23 · 707 HN pts
  3. CodeShow HN: Khoj – Chat offline with your second brain using Llama 2github.com · 2023-07-30 · 565 HN pts
  4. CodeM2 Ultra can run 128 streams of Llama 2 7B in parallelgithub.com · 2023-10-11
  5. ArticleGuide to running Llama 2 locallyreplicate.com · 2023-07-25 · 683 HN pts
  6. ArticleAccessing Llama 2 from the command-line with the LLM-replicate pluginsimonwillison.net · 2023-07-18 · 200 HN pts
  7. ArticleUnderstanding Llama 2 and the New Code Llama LLMsmagazine.sebastianraschka.com · 2023-08-30 · 170 HN pts
  8. CodeLlama 2 Everywhere (L2E): Standalone, Binary Portable, Bootable Llama 2github.com · 2023-10-05 · 320 HN pts
  9. CodeShow HN: LlamaGPT – Self-hosted, offline, private AI chatbot, powered by Llama 2github.com · 2023-08-16 · 225 HN pts
  10. PaperLlama 2 Longarxiv.org · 2023-09-29
  11. CodeLlama 2 on ONNX runs locallygithub.com · 2023-08-10
  12. PaperLlamaF: An Efficient Llama2 Architecture Accelerator on Embedded FPGAsarxiv.org · 2024-09-27
  13. ArticleLLaMA2 Chat 70B outperformed ChatGPTtatsu-lab.github.io · 2023-07-27
  14. ArticleFast and Portable Llama2 Inference on the Heterogeneous Edgesecondstate.io · 2023-11-13
  15. ArticleJetMoE: Reaching LLaMA2 performance with 0.1M dollarsresearch.myshell.ai · 2024-04-04
  16. CodeShow HN: Llama2 Embeddings FastAPI Servergithub.com · 2023-08-15
  17. ArticleFine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Custom Modelsanyscale.com · 2023-08-11
  18. ArticleA simple guide to fine-tuning Llama 2brev.dev · 2023-07-24
  19. ArticleWhat's new in Llama 2 and how to run it locallyagi-sphere.com · 2023-08-06
  20. DiscussionLeCun: Qualcomm working with Meta to run Llama-2 on mobile devicestwitter.com · 2023-07-23
  21. ArticleLoRA Fine-Tuning Efficiently Undoes Safety Training from Llama 2-Chat 70Blesswrong.com · 2023-10-13
  22. ArticleFast Llama 2 on CPUs with Sparse Fine-Tuning and DeepSparseneuralmagic.com · 2023-11-23
  23. DiscussionVicuna v1.5 series, featuring 4K and 16K context, based on Llama 2twitter.com · 2023-08-03

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 17 quantized builds and 100 Spaces.

  • As a service0no API provider listed
  • On own hardware17quantized builds
  • Built on6536 fine-tunes · 22 adapters · 7 merges
  • In apps100+Hugging Face Spaces

Top Spaces running it: Illia56/Ask-AI-Youtube ♥ 81 · EvanTHU/MotionLLM ♥ 43 · allenai/BaseChat ♥ 23 · allenai/URIAL-Bench ♥ 22

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 64 GB Mac (unified memory) at INT4

About 41 GB at INT4, 83 GB at FP8, 166 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 41 GB of weights (estimate; KV cache and activations extra)INT441 GBFP8: about 83 GB of weights (estimate; KV cache and activations extra)FP883 GBBF16: about 166 GB of weights (estimate; KV cache and activations extra)BF16166 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 41 GB of weights (estimate; KV cache and activations extra)INT441 GBFP8: about 83 GB of weights (estimate; KV cache and activations extra)FP883 GBBF16: about 166 GB of weights (estimate; KV cache and activations extra)BF16166 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

10 dated articles and 0 repo commits so far

Teal is the weights (2023-07-11), amber triangles are 10 dated articles (each links to it).

AugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctLlama 2 · 2023-07-18Llama2.c: Inference llama 2 in one file of pure C · 2023-07-23Show HN: Khoj – Chat offline with your second brain using Llama 2 · 2023-07-30Fine-tune your own Llama 2 to replace GPT-3.5/4 · 2023-09-12Run Llama 2 uncensored locally · 2023-08-02Guide to running Llama 2 locally · 2023-07-25Accessing Llama 2 from the command-line with the LLM-replicate plugin · 2023-07-18Understanding Llama 2 and the New Code Llama LLMs · 2023-08-30Llama 2 Everywhere (L2E): Standalone, Binary Portable, Bootable Llama 2 · 2023-10-05Show HN: LlamaGPT – Self-hosted, offline, private AI chatbot, powered by Llama 2 · 2023-08-16Weights 2023-07-11 · 2023-07-11Weights 2023-07-11now
AugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctLlama 2 · 2023-07-18Llama2.c: Inference llama 2 in one file of pure C · 2023-07-23Show HN: Khoj – Chat offline with your second brain using Llama 2 · 2023-07-30Fine-tune your own Llama 2 to replace GPT-3.5/4 · 2023-09-12Run Llama 2 uncensored locally · 2023-08-02Guide to running Llama 2 locally · 2023-07-25Accessing Llama 2 from the command-line with the LLM-replicate plugin · 2023-07-18Understanding Llama 2 and the New Code Llama LLMs · 2023-08-30Llama 2 Everywhere (L2E): Standalone, Binary Portable, Bootable Llama 2 · 2023-10-05Show HN: LlamaGPT – Self-hosted, offline, private AI chatbot, powered by Llama 2 · 2023-08-16Weights 2023-07-11 · 2023-07-11Weights 2023-07-11now

What people built on it

82 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

82 derivatives on the Hub, led by 36 fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

17 quantized (Hugging Face model tree)36 fine-tunes (Hugging Face model tree)7 merges (Hugging Face model tree)22 adapters (Hugging Face model tree)17 quantized36 fine-tunes7 merges22 adapters
17 quantized (Hugging Face model tree)36 fine-tunes (Hugging Face model tree)7 merges (Hugging Face model tree)22 adapters (Hugging Face model tree)17 quantized36 fine-tunes7 merges22 adapters
GGUF TheBloke/Llama-2-70B-GGUF for llama.cpp, Ollama, LM Studio 66.2k downloads GPTQ TheBloke/Llama-2-70B-GPTQ for vLLM, SGLang 374.5k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Facts

17 fields

Listed in LLMs

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesLlamaForCausalLM
Model typellama
Params by dtypeF32: 5,120 · F16: 68,976,648,192
Size bandXL
MoEno
Gatedmanual
PapersarXiv:2307.09288
LicenseLlama 2 Community License (llama2)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · LlamaForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · LlamaForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Llama-2-70b-hf: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)166 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)83 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)41 GB · Single high-end GPU (<80 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: meta-llama/Llama-2 82 derivatives on the Hub

quantized: 17finetune: 36adapter: 22merge: 7

Top variants (40 shown, ranked by downloads)

Relation
Quant
Size
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
TheBloke/Llama-2-70B-GPTQ quantized GPTQ 69.0B XL llama2 TheBloke 83 374.5k
allenai/tulu-2-dpo-70b finetune — 69.0B XL other allenai 156 83.9k
TheBloke/Llama-2-70B-AWQ quantized AWQ 69.0B XL llama2 TheBloke 14 79.9k
TheBloke/Llama-2-70B-GGUF quantized GGUF — — llama2 TheBloke 30 66.2k
karakuri-ai/karakuri-lm-70b-v0.1 finetune — 69.2B XL other karakuri-ai 27 23.6k
allenai/tulu-2-70b finetune — — — — allenai 8 10.5k
NousResearch/Nous-Hermes-2-Llama-2-70B finetune — — — apache-2.0 NousResearch 19 6k
eci-io/climategpt-70b finetune — 69.0B XL other eci-io 13 4.6k
facebook/layerskip-llama2-70B finetune — 69.0B XL other facebook 5 3.6k
allenai/tulu-v1-llama2-70b finetune — — — — allenai 1 3.4k
dfurman/Llama-2-70B-Instruct-v0.1 adapter — — — llama2 dfurman 14 1.7k
nvidia/OpenMath-Llama-2-70b-hf finetune — 69.0B XL llama2 nvidia 4 1.9k
donghyunli/Llama-2-70b-KronQ-W4A16-g128 finetune -W4A16 35.9B XL llama2 donghyunli 0 815
donghyunli/Llama-2-70b-KronQ-W3A16-g128-fake finetune — 69.0B XL llama2 donghyunli 0 716
donghyunli/Llama-2-70b-KronQ-W3A16-fake finetune — 69.0B XL llama2 donghyunli 0 711
AstroMLab/astrollama-2-70b-base_aic finetune — — — mit AstroMLab 2 397
donghyunli/Llama-2-70b-KronQ-W2A16-g128 finetune — 18.7B L llama2 donghyunli 0 494
AstroMLab/astrollama-2-70b-chat_aic finetune — — — mit AstroMLab 1 368
LoneStriker/tulu-2-70b-6.0bpw-h6-exl2 finetune 0bpw — — — LoneStriker 1 257
LoneStriker/tulu-2-dpo-70b-4.0bpw-h6-exl2 finetune 0bpw — — — LoneStriker 1 248
allenai/tulu-v2-qlora-70b adapter — — — — allenai 1 240
nvidia/OpenMath-Llama-2-70b finetune — — — llama2 nvidia 5 36
jamesdborin/llama2-70b-base-4bit-AWQ quantized -4bit 69.0B XL llama2 jamesdborin 0 269
LoneStriker/tulu-2-70b-2.4bpw-h6-exl2 finetune 4bpw — — — LoneStriker 0 242
LoneStriker/tulu-2-dpo-70b-2.65bpw-h6-exl2 finetune 5bpw — — — LoneStriker 0 231
from-our-page/BigLlama-2-120B finetune — 120B XXL — from-our-page 0 227
MatrixC7/karakuri-lm-70b-v0.1-exl2 finetune exl2 — — cc-by-sa-4.0 MatrixC7 0 211
LoneStriker/tulu-2-70b-5.0bpw-h6-exl2 finetune 0bpw — — — LoneStriker 0 209
BigHuggyD/jukofyork_miquplus-midnight-70b_exl2_5.0bpw merge exl2 — — other BigHuggyD 0 207
jusjinuk/Llama-2-70b-hf-4bit-GuidedQuant-LNQ quantized -4bit — — llama2 jusjinuk 0 200
iamshnoo/alpaca-2-70b-english adapter — — — cc-by-4.0 iamshnoo 0 195
ChuckMcSneed/Xwin-LM-70B-V0.1-LORA adapter — — — — ChuckMcSneed 1 144
brain-lab/Llama-2-70b-QuIP-HARP-2Bit finetune — 3.28B M llama2 brain-lab 0 189
iamshnoo/alpaca-2-70b-hindi adapter — — — cc-by-4.0 iamshnoo 0 177
LoneStriker/tulu-2-dpo-70b-5.0bpw-h6-exl2 finetune 0bpw — — — LoneStriker 2 75
haritzpuerto/LLaMA2-70B-dcot adapter — — — apache-2.0 haritzpuerto 0 172
BigHuggyD/jukofyork_miquplus-midnight-70b_exl2_7.5bpw merge exl2 — — other BigHuggyD 0 166
iamshnoo/alpaca-2-70b-swahili adapter — — — cc-by-4.0 iamshnoo 0 161
BigHuggyD/jukofyork_miquplus-midnight-70b_exl2_6.5bpw merge exl2 — — other BigHuggyD 0 159
BigHuggyD/jukofyork_miquplus-midnight-70b_exl2_6.0bpw merge exl2 — — other BigHuggyD 0 154

Articles and coverage

29 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Llama 2
ai.meta.com2023-07-18officialauthorityHN front pageabout the Llama 2 familyHN 2,268 pts · 820 comments42 words

Analysis8

Llama2.c: Inference llama 2 in one file of pure C
github.com2023-07-23authorityHN front pageabout the Llama 2 familyHN 707 pts · 165 comments4,774 words
The llama2.c repository offers a 700-line pure C file for training and inferring Llama 2 models, currently supporting fp32 inference for models up to 7B parameters. Work on model quantization is currently ongoing, but the project includes a 110M parameter model trained on the TinyStories dataset.written, reviewed, cited
  • Models larger than 7B are unusably slow because the repo performs float32 inference.
  • The 13B+ model variant fails due to integer flow errors in pointer arithmetic.
Show HN: Khoj – Chat offline with your second brain using Llama 2
github.com2023-07-30authorityHN front pageabout the Llama 2 familyHN 565 pts · 150 comments218 words
Khoj is an open-source, self-hostable AI platform that lets users get answers from the web or their docs. It enables building custom agents and scheduling automations while supporting online or local LLMs like gpt, claude, gemini, llama, qwen, and mistral. Users can get started for free and contribute to the community-sustained project.written, reviewed, cited
Run Llama 2 uncensored locally
ollama.ai2023-08-02HN front pageabout the Llama 2 familyHN 722 pts · 212 comments1,251 words
Ollama enables local execution of uncensored Llama 2 variants, such as the 7B model fine-tuned using Wizard-Vicuna conversation dataset and Nous Hermes Llama 2 13B fine-tuned on over 300,000 instructions. The 13B model stands out for its long responses, lower hallucination rate, and absence of OpenAI censorship mechanisms. Users can run these models via commands like ollama run llama2-uncensored to bypass standard safety refusals.written, reviewed, cited
  • The uncensored model provides a specific recipe for dangerously spicy mayo including ingredients like cayenne pepper and paprika.
  • The uncensored model describes the manufacturing process for Tylenol by combining paracetamol with aspirin, caffeine, and diphenhydramine.
Guide to running Llama 2 locally
replicate.com2023-07-25HN front pageabout the Llama 2 familyHN 683 pts · 170 comments603 words
Accessing Llama 2 from the command-line with the LLM-replicate plugin
simonwillison.net2023-07-18authorityHN front pageabout the Llama 2 familyHN 200 pts · 43 comments1,144 words
Understanding Llama 2 and the New Code Llama LLMs
magazine.sebastianraschka.com2023-08-30authorityHN front pageabout the Llama 2 familyHN 170 pts · 34 comments4,048 words
Llama 2 Everywhere (L2E): Standalone, Binary Portable, Bootable Llama 2
github.com2023-10-05authorityHN front pageabout the Llama 2 familyHN 320 pts · 55 comments4,153 words
Show HN: LlamaGPT – Self-hosted, offline, private AI chatbot, powered by Llama 2
github.com2023-08-16authorityHN front pageabout the Llama 2 familyHN 225 pts · 75 comments

Community1

Fine-tune your own Llama 2 to replace GPT-3.5/4
news.ycombinator.com2023-09-12HN front pageabout the Llama 2 familyHN 955 pts · 181 comments10,023 words
Fine-tuning open-source Llama 2 models can replace GPT-3.5 and GPT-4 for specific tasks, with a fine-tuned Llama 7B matching GPT-4 labels 95% of the time on a test set. While fine-tuning requires 1000+ examples and is less flexible than prompting, a fine-tuned Llama 7B is 50x cheaper than GPT-3.5 per token, costing just $19 to classify 2M recipes versus $23k for GPT-4.written, reviewed, cited
All coverage tracked for this model (29) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
LlamaF: An Efficient Llama2 Architecture Accelerator on Embedded FPGAsarxiv.org2024-09-272026-10-07hn
JetMoE: Reaching LLaMA2 performance with 0.1M dollarsresearch.myshell.ai2024-04-042026-10-07hn
Fast Llama 2 on CPUs with Sparse Fine-Tuning and DeepSparseneuralmagic.com2023-11-232026-10-07hn
Fast and Portable Llama2 Inference on the Heterogeneous Edgesecondstate.io2023-11-132026-10-07hn
LoRA Fine-Tuning Efficiently Undoes Safety Training from Llama 2-Chat 70Blesswrong.com2023-10-132026-10-07hn
M2 Ultra can run 128 streams of Llama 2 7B in parallelgithub.com2023-10-112026-10-07hn
Llama 2 Everywhere (L2E): Standalone, Binary Portable, Bootable Llama 2github.com2023-10-052026-10-07hn
Llama 2 Longarxiv.org2023-09-292026-10-07hn
Fine-tune your own Llama 2 to replace GPT-3.5/4news.ycombinator.com2023-09-122026-10-07hn
Understanding Llama 2 and the New Code Llama LLMsmagazine.sebastianraschka.com2023-08-302026-10-07hn
Llama 2 is about as factually accurate as GPT-4 for summaries and is 30X cheaperanyscale.com2023-08-292026-10-07hn
WebLLM: Llama2 in the Browserwebllm.mlc.ai2023-08-292026-10-07hn
Show HN: LlamaGPT – Self-hosted, offline, private AI chatbot, powered by Llama 2github.com2023-08-162026-10-07hn
Show HN: Llama2 Embeddings FastAPI Servergithub.com2023-08-152026-10-07hn
Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Custom Modelsanyscale.com2023-08-112026-10-07hn
Llama 2 on ONNX runs locallygithub.com2023-08-102026-10-07hn
Using LLama2 to find links in mental disorders and ideological fixationmindgrid.io2023-08-092026-10-07hn
What's new in Llama 2 and how to run it locallyagi-sphere.com2023-08-062026-10-07hn
Vicuna v1.5 series, featuring 4K and 16K context, based on Llama 2twitter.com2023-08-032026-10-07hn
Cookbook: Finetuning Llama 2 in your own cloud environment, privatelyblog.skypilot.co2023-08-022026-10-07hn
Run Llama 2 uncensored locallyollama.ai2023-08-022026-10-07hn
Show HN: Khoj – Chat offline with your second brain using Llama 2github.com2023-07-302026-10-07hn
LLaMA2 Chat 70B outperformed ChatGPTtatsu-lab.github.io2023-07-272026-10-07hn
Guide to running Llama 2 locallyreplicate.com2023-07-252026-10-07hn
A simple guide to fine-tuning Llama 2brev.dev2023-07-242026-10-07hn
Llama2.c: Inference llama 2 in one file of pure Cgithub.com2023-07-232026-10-07hn
LeCun: Qualcomm working with Meta to run Llama-2 on mobile devicestwitter.com2023-07-232026-10-07hn
Accessing Llama 2 from the command-line with the LLM-replicate pluginsimonwillison.net2023-07-182026-10-07hn
Llama 2ai.meta.com official2023-07-182026-10-07hn

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Similar models

3 models

Similar models

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:09 UTC · Last verified 2026-10-25

Follow Llama-2-70b-hf

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.