Model Follower.

Up to date

Rebuilt
2026-10-26 07:36 UTC
Release radar
116 on watch
Hub snapshot
2026-10-26
Re-checked
daily · every number sourced and dated

Models › granite-3.3 family

granite-3.3-8b-instruct

Hugging Face ↗ Follow changes

ibm-granite Apache 2.0 out 19 months ago

8.17B parameters, text in, text out, 128K tokens of context.

Verified 2026-10-26 07:27 UTC Rank #711 in LLMs · score 72 method v1 Sources 3 Downloads 30d 69.4k · #354 of 619 tracked

Can I run it?4.9GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-26
Do people use it?69.4kdownloads, 30 days#354 of 619 we track · below the median+0 likes, downloads −19% this week161 likesHugging Face2026-10-26

Worth reading

Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.

Latest · 18 months ago

IBM Granite 3.3: Speech recognition, refined reasoning, and RAG LoRAs

IBM releases Granite Speech 3.3 8B, a speech-to-text model built on Granite 3.3 8B Instruct that processes a 20-minute audio file on an H100 80GB GPU.

  • Granite Speech 3.3 uses a two-pass design requiring an initial transcription call before querying the model about the audio content.
  • Although the model can process 20-minute files, IBM suggests a 1-minute limit per discrete unit to maintain consistent accuracy.
ibm.com 2025-04-17 3 HN pts 2.5k words about the granite 3.3 family family

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: 33 quantized builds and 14 Spaces.

  • As a service0no API provider listed
  • On own hardware33quantized builds
  • Built on2917 fine-tunes · 12 adapters · 0 merges
  • In apps14Hugging Face Spaces

Top Spaces running it: responsible-prompting/demo-multiturn ♥ 7 · responsible-prompting/demo ♥ 3 · frozen8569/sahay-ai ♥ 1

Will it fit?

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

Fits a 16 GB card at INT4

About 4.9 GB at INT4, 9.8 GB at FP8, 20 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 4.9 GB of weights (estimate; KV cache and activations extra)INT44.9 GBFP8: about 9.8 GB of weights (estimate; KV cache and activations extra)FP89.8 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 4.9 GB of weights (estimate; KV cache and activations extra)INT44.9 GBFP8: about 9.8 GB of weights (estimate; KV cache and activations extra)FP89.8 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 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.

1 dated article and 21 repo commits so far

Teal is the weights (2025-04-09), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.

MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-05-12: 1 repo commit2025-05-08: 1 repo commit2025-05-02: 1 repo commit2025-04-24: 1 repo commit2025-04-16: 3 repo commits2025-04-14: 2 repo commits2025-04-11: 6 repo commits2025-04-10: 4 repo commits2025-04-09: 2 repo commitsIBM Granite 3.3: Speech recognition, refined reasoning, and RAG LoRAs · 2025-04-17Weights 2025-04-09 · 2025-04-09Weights 2025-04-09now
MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-05-12: 1 repo commit2025-05-08: 1 repo commit2025-05-02: 1 repo commit2025-04-24: 1 repo commit2025-04-16: 3 repo commits2025-04-14: 2 repo commits2025-04-11: 6 repo commits2025-04-10: 4 repo commits2025-04-09: 2 repo commitsIBM Granite 3.3: Speech recognition, refined reasoning, and RAG LoRAs · 2025-04-17Weights 2025-04-09 · 2025-04-09Weights 2025-04-09now

What people built on it

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

62 derivatives on the Hub, most of them quantized

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

33 quantized (Hugging Face model tree)17 fine-tunes (Hugging Face model tree)12 adapters (Hugging Face model tree)33 quantized17 fine-tunes12 adapters
33 quantized (Hugging Face model tree)17 fine-tunes (Hugging Face model tree)12 adapters (Hugging Face model tree)33 quantized17 fine-tunes12 adapters
GGUF ibm-granite/granite-3.3-8b-instruct-GGUF for llama.cpp, Ollama, LM Studio 171.6k downloads MLX mlx-community/granite-3.3-8b-instruct-4bit for Apple silicon 2.7k downloads GPTQ jeffcookio/granite-3.3-8b-instruct-gptqmodel-4b-64g for vLLM, SGLang 217.6k 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 released under the Apache 2.0 license and supports a 128K context length.[1][2]

  1. license
  2. “128K context length”

Generated 2026-10-07 15:14 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

6 use cases

What it is good at 6 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
ArchitecturesGraniteForCausalLM
Model typegranite
Dtypebfloat16
Params by dtypeBF16: 8,170,864,640
Size bandM
MoEno
Context length131,072 (config.json max_position_embeddings)
PapersarXiv:0000.00000
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-26

Servability

vLLM · SGLang

Servability

vLLM: supported · GraniteForCausalLM · 4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60SGLang: supported · GraniteForCausalLM · 60d941293e3af27a12c6eb64e2c3e2e1a17f100f

vLLM registry (4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60) · SGLang registry (60d941293e3af27a12c6eb64e2c3e2e1a17f100f) · 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 granite-3.3-8b-instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)20 GB · Single GPU (<24 GB)
FP8 (1 byte/param)9.8 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)4.9 GB · Single GPU (<24 GB)

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

All variants

40 variants

Lineage

Fine-tuned from ibm-granite/granite-3.3-8b-base (same organization) Family: ibm-granite/granite-3.3 62 derivatives on the Hub

Base model: ibm-granite/granite-3.3-8b-base

finetune: 17quantized: 33adapter: 12

Top variants (40 shown, ranked by downloads)

Relation
Quant
Size
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
ibm-granite/granite-speech-3.3-8b finetune — 8.65B M apache-2.0 ibm-granite 171 880.1k
ibm-ai-platform/micro-g3.3-8b-instruct-1b finetune — 1.20B S apache-2.0 ibm-ai-platform 6 290.7k
jeffcookio/granite-3.3-8b-instruct-gptqmodel-4b-64g quantized gptq 8.17B M — jeffcookio 1 217.6k
ibm-granite/granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 ibm-granite 40 171.6k
ibm-granite/granite-3.3-8b-instruct-FP8 quantized -FP8 8.17B M apache-2.0 ibm-granite 3 63.6k
unsloth/granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 unsloth 7 30k
bartowski/ibm-granite_granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 bartowski 3 24.8k
Raymond-dev-546730/ClinicalThought-AI-8B adapter GGUF — — apache-2.0 Raymond-dev-546730 11 7.6k
mradermacher/granite-3.3-8b-instruct-i1-GGUF quantized GGUF — — apache-2.0 mradermacher 0 7k
lmstudio-community/granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 lmstudio-community 0 4.9k
ibm-granite/granite-3.3-8b-math-prm-v2 finetune — 8.17B M apache-2.0 ibm-granite 14 2.8k
mradermacher/granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 mradermacher 0 3.4k
mlx-community/granite-3.3-8b-instruct-4bit quantized -4bit 8.17B M apache-2.0 mlx-community 1 2.7k
mlx-community/granite-3.3-8b-instruct-fp16 finetune -fp16 8.17B M apache-2.0 mlx-community 3 2.3k
drawais/Granite-3.3-8B-Instruct-AWQ-INT4 quantized AWQ 8.23B M apache-2.0 drawais 0 2.2k
DevQuasar-12/ibm-granite.granite-3.3-8b-instruct-GGUF quantized GGUF — — — DevQuasar-12 0 2.1k
Qiskit/granite-3.3-8b-qiskit finetune — 8.17B M apache-2.0 Qiskit 7 1.2k
mlx-community/granite-3.3-8b-instruct-8bit quantized -8bit 8.17B M apache-2.0 mlx-community 2 903
AbteeXAILab/lumynax-infused-granite33-8b-gguf quantized gguf — — apache-2.0 AbteeXAILab 0 769
ibm-granite/granite-3.3-8b-rag-agent-lib finetune — — — apache-2.0 ibm-granite 13 0
lucas-sousa-pereira/granite-3-3-8b-java-code-refactoring adapter GGUF — — apache-2.0 lucas-sousa-pereira 0 638
TevunahAi/Granite-3.3-8B-Instruct-GPTQ quantized GPTQ 8.17B M apache-2.0 TevunahAi 0 607
ibm-granite/granite-3.3-8b-lora-math-prm finetune — — — apache-2.0 ibm-granite 10 0
smarttasks/granite-3.3-8b-instruct-GGUF quantized GGUF — — apache-2.0 smarttasks 0 432
mlx-community/granite-3.3-8b-instruct-6bit quantized — 8.17B M apache-2.0 mlx-community 1 344
Artificial-Development-Studios/QuantumAI-granite-3.3-8b-qiskit finetune — 8.17B M apache-2.0 Artificial-Development-Studios 1 329
NikolayKozloff/granite-3.3-8b-instruct-Q8_0-GGUF quantized Q8_0 — — apache-2.0 NikolayKozloff 2 252
stelterlab/granite-3.3-8b-instruct-FP8-Dynamic quantized -FP8 8.37B M apache-2.0 stelterlab 0 325
ibm-granite/granite-3.3-8b-security-lib finetune — — — apache-2.0 ibm-granite 6 0
Fmuaddib/granite-3.3-8b-instruct-mlx-fp16 finetune mlx 8.17B M apache-2.0 Fmuaddib 0 298
Fmuaddib/granite-3.3-8b-instruct-mlx-8Bit quantized mlx 8.17B M apache-2.0 Fmuaddib 0 261
AmpereComputing/granite-3.3-8b-instruct-gguf quantized gguf — — — AmpereComputing 0 239
cgus/granite-3.3-8b-instruct-exl2 quantized exl2 — — apache-2.0 cgus 0 211
SandLogicTechnologies/Granite-3.3-8B-Instruct-GGUF quantized GGUF — — apache-2.0 SandLogicTechnologies 0 210
LogicBombaklot/granite-3.3-8b-instruct-mlx-8Bit quantized mlx 8.17B M apache-2.0 LogicBombaklot 0 200
heterodoxin/granite-3.3-8b-instruct-apostate finetune — 8.37B M — heterodoxin 1 139
alexj03/granite-3.3-8b-instruct-abliterated finetune — 8.17B M apache-2.0 alexj03 1 93
cobrakenji/granite-3.3-8b-instruct-Q4_K_M-GGUF quantized GGUF — — apache-2.0 cobrakenji 0 131
vpakarinen/ibm-granite-8b-lora-uncensored adapter — — — apache-2.0 vpakarinen 0 126
pasan1992/granite-3.3-8b-instruct-Q4_K_M-GGUF quantized GGUF — — apache-2.0 pasan1992 0 126

Repository changelog

21 commits

Repository changelog 21 commits

DateCommitSHA
2025-05-12 Fix typo (#7) 51dd4bc
2025-05-08 update to resources link 041d6ba
2025-05-02 Update README.md 8fe5f09
2025-04-24 Removed extra space and added a new line to chat template (#5) 838cee3
2025-04-16 Update README.md 115e19f
2025-04-16 Update README.md b318865
2025-04-16 Update README.md 85233b2
2025-04-14 Fix use_cache to true (#2) de4d392
2025-04-14 Update README.md e693106
2025-04-11 Update README.md dd6c8af
2025-04-11 Update README.md b9a7d27
2025-04-11 Update README.md dba65d4
2025-04-11 Update README.md 3c039d1
2025-04-11 Update README.md cefe3ca
2025-04-11 Update README.md e2b849f
2025-04-10 Support `tools` as input alongside `available_tools` (#1) 2faede9
2025-04-10 Update README.md 9cb34ad
2025-04-10 Update README.md ea56e83
2025-04-10 3.3-8b-instruct model chat template change 6e089e6
2025-04-09 3.3-8b-instruct model weights upload 3efd179
2025-04-09 initial commit 4c29036

Articles and coverage

1 piece

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

IBM Granite 3.3: Speech recognition, refined reasoning, and RAG LoRAs
ibm.com2025-04-17officialauthorityabout the granite 3.3 familyHN 3 pts · 0 comments2,515 words
IBM releases Granite Speech 3.3 8B, a speech-to-text model built on Granite 3.3 8B Instruct that processes a 20-minute audio file on an H100 80GB GPU. The Granite 3.3 Instruct models now offer fill-in-the-middle (FIM) capabilities in addition to standard next-token prediction. IBM Research has also developed a series of activated LoRAs (aLoRAs) alongside conventional RAG-focused LoRA adapters for Granite 3.2.written, reviewed, cited
  • Granite Speech 3.3 uses a two-pass design requiring an initial transcription call before querying the model about the audio content.
  • Although the model can process 20-minute files, IBM suggests a 1-minute limit per discrete unit to maintain consistent accuracy.
  • Activated LoRAs (aLoRAs) reuse the existing key-value cache to avoid recomputing context, unlike standard LoRAs that slow performance.

Similar models

3 models

Similar models

  • Llama-3.1-8B-Instruct · 8.03B
    same size band (M), same task (text-generation), more providers
  • Qwen3-8B · 8.19B
    same size band (M), same task (text-generation), more providers, permissive license
  • Meta-Llama-3-8B-Instruct · 8.03B
    same size band (M), same task (text-generation), more providers

Sources

3 sources

Sources

Hub snapshot: 2026-10-26 · HF API: 2026-10-26 07:24 UTC · Providers: 2026-10-26 07:26 UTC · Articles: 2026-10-26 07:27 UTC · Summary: 2026-10-07 15:14 UTC · Last verified 2026-10-26 07:27 UTC

Follow granite-3.3-8b-instruct

Watch this model

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