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 › bloomz family

bloomz-560m

Hugging Face ↗ Follow changes

bigscience BLOOM RAIL 1.0 out 4 years ago

559M parameters, text in, text out, 2K tokens of context.

Verified 2026-10-25 Rank #238 in LLMs · score 86.5 method v1 Sources 3 Downloads 30d 802.3k · #131 of 619 tracked

Can I run it?0.3GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?BLOOM RAIL 1.0Attribution requiredlicence file2026-10-24
Do people use it?802.3kdownloads, 30 days#131 of 619 we track · top 21%+2 likes, downloads −36% this week149 likesHugging Face2026-10-24

Worth reading

Picked by rule from 5 tracked pieces: the lab's own post, the most shared, a field report, the latest.

5 best of 5 tracked

The 1 pick above, plus these 4. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. CodeBloomz.cpp: Run multilingual BLOOM model with C++github.com · 2023-03-16 · 5 HN pts
  2. ArticleBLOOMZ 560M – 2k context, open weights | LLM Referencellmreference.com · undated
  3. ArticleScaling of Distributed Inference of BLOOMZ-560M Lightweight Large Language Models using Petals Tool : Analyzing Response Time, Resource Usage and Scaling Limits on CPU Infrastructureexa.ai · 2026-01-01
  4. Articlebloomz — VRAM, Quants & GPU Compatibilitynodepedia.com · 2022-09-17

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: 7 quantized builds and 100 Spaces.

  • As a service0no API provider listed
  • On own hardware7quantized builds
  • Built on31233 fine-tunes · 279 adapters · 0 merges
  • In apps100+Hugging Face Spaces

Top Spaces running it: Sharathhebbar24/One-stop-for-Open-source-models ♥ 96 · KBaba7/Quant ♥ 51 · FallnAI/Quantize-HF-Models ♥ 19 · bhaskartripathi/LLM_Quantization ♥ 16

Will it fit?

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

Fits a 16 GB card at INT4

About 0.3 GB at INT4, 0.7 GB at FP8, 1.3 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.3 GB of weights (estimate; KV cache and activations extra)INT40.3 GBFP8: about 0.7 GB of weights (estimate; KV cache and activations extra)FP80.7 GBBF16: about 1.3 GB of weights (estimate; KV cache and activations extra)BF161.3 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.3 GB of weights (estimate; KV cache and activations extra)INT40.3 GBFP8: about 0.7 GB of weights (estimate; KV cache and activations extra)FP80.7 GBBF16: about 1.3 GB of weights (estimate; KV cache and activations extra)BF161.3 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.

4 dated articles and 13 repo commits so far

Teal is the weights (2022-10-08), amber triangles are 4 dated articles (each links to it), grey ticks are repo commits.

OctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2023-05-27: 1 repo commit2023-03-16: 1 repo commit2023-02-10: 2 repo commits2022-12-03: 1 repo commit2022-11-18: 1 repo commit2022-11-07: 1 repo commit2022-11-04: 2 repo commits2022-10-09: 2 repo commits2022-10-08: 2 repo commitsBloomz.cpp: Run multilingual BLOOM model with C++ · 2023-03-16Run Language Models in pure C++: Introducing bloomz.cpp · 2023-03-16Scaling of Distributed Inference of BLOOMZ-560M Lightweight Large Language Models using Petals Tool : Analyzing Response Time, Resource Usage and Scaling Limits on CPU Infrastructure · 2026-01-01bloomz — VRAM, Quants & GPU Compatibility · 2022-09-17Weights 2022-10-08 · 2022-10-08Weights 2022-10-08now
OctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2023-05-27: 1 repo commit2023-03-16: 1 repo commit2023-02-10: 2 repo commits2022-12-03: 1 repo commit2022-11-18: 1 repo commit2022-11-07: 1 repo commit2022-11-04: 2 repo commits2022-10-09: 2 repo commits2022-10-08: 2 repo commitsBloomz.cpp: Run multilingual BLOOM model with C++ · 2023-03-16Run Language Models in pure C++: Introducing bloomz.cpp · 2023-03-16Scaling of Distributed Inference of BLOOMZ-560M Lightweight Large Language Models using Petals Tool : Analyzing Response Time, Resource Usage and Scaling Limits on CPU Infrastructure · 2026-01-01bloomz — VRAM, Quants & GPU Compatibility · 2022-09-17Weights 2022-10-08 · 2022-10-08Weights 2022-10-08now

What people built on it

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

319 derivatives on the Hub, most of them adapters

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

7 quantized (Hugging Face model tree)33 fine-tunes (Hugging Face model tree)279 adapters (Hugging Face model tree)7 quantized33 fine-tunes279 adapters
7 quantized (Hugging Face model tree)33 fine-tunes (Hugging Face model tree)279 adapters (Hugging Face model tree)7 quantized33 fine-tunes279 adapters
GGUF mradermacher/bloomz-560m-i1-GGUF for llama.cpp, Ollama, LM Studio 3.9k downloads

All variants →

Reference

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

What the card says

1 cited sentence

Summary generated, cited

The model is part of a family capable of following human instructions in dozens of languages zero-shot.[1]

  1. “capable of following human instructions in dozens of lang…”

Generated 2026-10-07 15:05 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

4 use cases

What it is good at 4 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in LLMs

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesBloomForCausalLM
Model typebloom
Params by dtypeF16: 559,214,592
Size bandS
MoEno
Context length2,048 (config.json max_position_embeddings)
Languagesak, ar, as, bm, bn, ca, code, en, es, eu, fon, fr, gu, hi, id, ig, ki, kn, lg, ln, ml, mr, ne, nso, ny, or, pa, pt, rn, rw, sn, st, sw, ta, te, tn, ts, tum, tw, ur, vi, wo, xh, yo, zh, zu
PapersarXiv:2211.01786
LicenseBLOOM RAIL 1.0 (bigscience-bloom-rail-1.0)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · BloomForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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 bloomz-560m: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)1.3 GB · Edge (<4 GB)
FP8 (1 byte/param)0.7 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)0.3 GB · Edge (<4 GB)

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

All variants

40 variants

Lineage

Original model Family: bigscience/bloomz 319 derivatives on the Hub

quantized: 7finetune: 33adapter: 279

Top variants (40 shown, ranked by downloads)

Relation
Quant
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
mradermacher/bloomz-560m-i1-GGUF quantized GGUF — — bigscience-bloom-rail-1.0 mradermacher 0 3.9k
tensorblock/bloomz-560m-GGUF quantized GGUF — — bigscience-bloom-rail-1.0 tensorblock 0 3.6k
mradermacher/bloomz-560m-GGUF quantized GGUF — — bigscience-bloom-rail-1.0 mradermacher 0 3.1k
Xenova/bloomz-560m quantized — — — — Xenova 1 1.4k
liodon-ai/bloomz-560m-imatrix-GGUF quantized imatrix — — other liodon-ai 0 1.3k
tasal9/pashto-base-bloom finetune — 559M S mit tasal9 0 824
difince/bloomz-560m_PROMPT_TUNING_CAUSAL_LM adapter — — — — difince 0 608
saketh-gootykase/unh-academic-integrity-policy-560m finetune — 559M S bigscience-bloom-rail-1.0 saketh-gootykase 1 476
HARSHAPALNATIUNH/results finetune — 559M S bigscience-bloom-rail-1.0 HARSHAPALNATIUNH 0 517
HARSHAPALNATIUNH/Automodelbloom finetune — 559M S bigscience-bloom-rail-1.0 HARSHAPALNATIUNH 0 486
jysssacc/bloomz-560m_fine_lr0.0005_bs10_epoch5_wd0.01 finetune — 559M S bigscience-bloom-rail-1.0 jysssacc 0 377
jysssacc/bloomz-560m_fine_lr5e-05_bs10_epoch5_wd0.01 finetune — 559M S bigscience-bloom-rail-1.0 jysssacc 0 369
vj99/bloomz-560m_PROMPT_TUNING_CAUSAL_LM adapter — — — — vj99 0 351
error577/39d93b8a-13bf-40e9-8ba8-8d338e0337b1 adapter — — — bigscience-bloom-rail-1.0 error577 0 281
Neeraja333/llm-560m finetune — 559M S bigscience-bloom-rail-1.0 Neeraja333 0 279
intelia-lab-uah/bloomz-560m_QG_SS finetune — 559M S bigscience-bloom-rail-1.0 intelia-lab-uah 0 257
kalyanmaram/UNH-QA-bloom-560m finetune — 559M S bigscience-bloom-rail-1.0 kalyanmaram 0 245
intelia-lab-uah/bloomz-560m_QG_SQAC finetune — 559M S bigscience-bloom-rail-1.0 intelia-lab-uah 0 225
Luciano/lora-bloomz-560m-lener_br adapter — — — — Luciano 0 201
Luciano/qlora-bloomz-560m-lener_br adapter — — — — Luciano 0 188
intelia-lab-uah/bloomz-560m_QG_SQUAD finetune — 559M S bigscience-bloom-rail-1.0 intelia-lab-uah 0 185
intelia-lab-uah/bloomz-560m_AE_SQUAD finetune — 559M S bigscience-bloom-rail-1.0 intelia-lab-uah 0 182
intelia-lab-uah/bloomz-560m_AE_SS finetune — 559M S bigscience-bloom-rail-1.0 intelia-lab-uah 0 176
chitranshu324324/personality_llm_culture finetune — 559M S — chitranshu324324 1 114
NikoK/bloomz-560m_PROMPT_TUNING_CAUSAL_LM adapter — — — — NikoK 0 140
kalyanmaram/UNH-Academic-Integrity-bloomz-560m finetune — 559M S bigscience-bloom-rail-1.0 kalyanmaram 1 83
justtuananh/sentiment-prefix adapter — — — — justtuananh 0 128
KGsteven/bloomz-560m_PROMPT_TUNING_CAUSAL_LM_10000 adapter — — — — KGsteven 0 114
Luciano/ia3-bloomz-560m-lener_br adapter — — — — Luciano 0 110
rcarey/bloomz-560m_PROMPT_TUNING_CAUSAL_LM adapter — — — — rcarey 0 100
HARSHAPALNATIUNH/BLOOM560mfinetune finetune — 559M S bigscience-bloom-rail-1.0 HARSHAPALNATIUNH 0 96
HARSHAPALNATIUNH/Githubmodel finetune — 559M S bigscience-bloom-rail-1.0 HARSHAPALNATIUNH 0 96
shahzebnaveed/bloomz-560m_prompt_tuning_clm adapter — — — — shahzebnaveed 0 94
dada22231/9520e68d-45c4-4d41-be88-742e7e101bf8 adapter — — — bigscience-bloom-rail-1.0 dada22231 0 90
NIKHIL4053/fine_tuned_lora_model_1 adapter — — — — NIKHIL4053 0 85
kalyanmaram/UNH-Academic-Integrity-bloomz-560m-v4 finetune — 559M S bigscience-bloom-rail-1.0 kalyanmaram 0 84
jysssacc/bloomz-560m_fine_lr5e-05_bs4_epoch20_wd0.01 finetune — 559M S bigscience-bloom-rail-1.0 jysssacc 0 84
khalilUoM/bloom_p560m_5 finetune — — — bigscience-bloom-rail-1.0 khalilUoM 0 83
Alphatao/74ac04d8-7cae-4beb-9835-2a76ba1aeef3 finetune — — — — Alphatao 0 82
jysssacc/bloomz-560m_fine_lr5e-06_bs10_epoch5_wd0.01 finetune — 559M S bigscience-bloom-rail-1.0 jysssacc 0 78

Repository changelog

13 commits

Repository changelog 13 commits

DateCommitSHA
2023-05-27 Update README.md a2845d7
2023-03-16 Adding `safetensors` variant of this model (#7) 25f241f
2023-02-10 Fix architecture (#6) 64a6f17
2023-02-10 Fix architecture (#5) 842d973
2022-12-03 Update README.md e183ebb
2022-11-18 Update README.md 1bab45f
2022-11-07 Remove eval folder (Moved to bigscience/evaluation-results) 790efe9
2022-11-04 Add paper 63fdc61
2022-11-04 Create README.md e221b2d
2022-10-09 Organize 87637f3
2022-10-09 Add eval files cbda70b
2022-10-08 Add files 4e82575
2022-10-08 initial commit e245f9e

Articles and coverage

5 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.

Analysis4

Bloomz.cpp: Run multilingual BLOOM model with C++
github.com2023-03-16authorityabout the bloomz familyHN 5 pts · 2 comments656 words
A C++ implementation for BLOOM-like models is available in a GitHub repository built on top of llama.cpp. It supports all models loadable via BloomForCausalLM.from_pretrained() and allows conversion to the ggml format for inference. Users can convert the 7B1 model to ggml FP16 format and optionally quantize it to 4-bits. The bloomz-7b1-f16-q4_0 variant requires 5.3 GB of memory.written, reviewed, cited
  • The iOS proof-of-concept app requires a specific file named ggml-model-bloomz-560m-f16.bin to be placed in the Bloomer directory.
  • The 7B1 model loads with a context window of 512 tokens and uses 30 layers with 32 heads.
  • Inference on the 7B1 model consumes 5312.64 MB of context size and 480.00 MB of memory size.
BLOOMZ 560M – 2k context, open weights | LLM Reference
llmreference.com231 words
BLOOMZ 560M is a multilingual large language model released on 2022-07-20 with open weights and 2K context. Developed by fine-tuning BLOOM and mT5 models on a cross-lingual dataset, it exhibits strong cross-lingual generalization through zero-shot learning. The model houses 560 million parameters and is licensed under bigscience-bloom-rail-1.0.written, reviewed, cited
Scaling of Distributed Inference of BLOOMZ-560M Lightweight Large Language Models using Petals Tool : Analyzing Response Time, Resource Usage and Scaling Limits on CPU Infrastructure
exa.ai2026-01-01337 words
Distributed inference of the BLOOMZ-560M model using Petals on CPU infrastructure shows scaling is strongly strategy-dependent. Pipelined execution achieved the highest gains, sustaining a mean speedup of ≈ 1.6× at N = 10, but with a 13× increase in network footprint. Sequential execution degraded monotonically with scale, while microbatched execution exhibited stability only at small cluster sizes (N ≤ 3).written, reviewed, cited
bloomz — VRAM, Quants & GPU Compatibility
nodepedia.com2022-09-17about the bloomz family633 words
BLOOMZ is a 176.25B parameter model by BigScience released 2022-09-17 licensed under bigscience-bloom-rail-1.0 that speaks dozens of languages natively, unlike newer English-first models. It requires a minimum of 116GB VRAM for Q4_K_M quantization, supports a 2K context window, and aggressive 4-bit quant fits on one high-end datacenter GPU. The stated VRAM figures cover model weights plus a runtime allowance of about 10% but do not include the KV cache.written, reviewed, cited
  • BLOOMZ instruction-following is less polished than newer models like Llama and Mistral that came after it.
  • The aggressive 4-bit quantization softens quality noticeably while bringing the model to a single high-end datacenter GPU.
  • BLOOMZ handles light code generation but does not shine against code-specialized peers.

Community1

Run Language Models in pure C++: Introducing bloomz.cpp
twitter.com2023-03-16about the bloomz familyHN 1 pts · 0 comments

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

  • bloom-560m · 559M
    same size band (S), same task (text-generation)
  • Qwen2.5-0.5B-Instruct · 494M
    same size band (S), same task (text-generation), more providers, permissive license
  • Qwen3-0.6B-Base · 596M
    same size band (S), same task (text-generation), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:37 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:05 UTC · Last verified 2026-10-25

Follow bloomz-560m

Watch this model

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