SAVRN
Search Contact SAVRN

Open-weight model

kuza-qwen-3.5-4b

by Kuza AI kuzaai/kuza-qwen-3.5-4b

Full training-run archive for Kuza (East Africa agricultural assistant), fine-tuned from unsloth/Qwen3.5-4B. Weights, logs, checkpoints, GGUFs, and provenance are stored with the same layout as $KUZAWORKDIR/kuza-qwen-3.5-4b/.

Parameters
Context
Weights29.0 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads

Model Card

By Kuza AI, published under apache-2.0, revision 9c2d90a8f6a3.

Full training-run archive for Kuza (East Africa agricultural assistant), fine-tuned from unsloth/Qwen3.5-4B. Weights, logs, checkpoints, GGUFs, and provenance are stored with the same layout as $KUZAWORKDIR/kuza-qwen-3.5-4b/. This derivative is subject to the Apache 2.0 license of the Qwen base model. - 100% English train from kuzaai/kuzasftenglish - 35% Swahili from kuzaai/kuzasftswahili - 8% HuggingFaceH4/norobots - 5% adversarial from kuzaai/kuzasftadversarial - all multiturn from kuzaai/kuzasftmultiturn Sequence length 1024, 2 epochs, LR 0.0002. Thinking is off (--reasoning off when llama.cpp supports it). GGUFs are text-only (MTP/nextn kept; vision/audio dropped). ssmout is Q6K on…

Read Kuza AI's full model card

Full training-run archive for Kuza (East Africa agricultural assistant), fine-tuned from unsloth/Qwen3.5-4B. Weights, logs, checkpoints, GGUFs, and provenance are stored with the same layout as $KUZA_WORK_DIR/kuza-qwen-3.5-4b/.

This derivative is subject to the Apache 2.0 license of the Qwen base model.

Training mix

  • 100% English train from kuzaai/kuza_sft_english
  • 35% Swahili from kuzaai/kuza_sft_swahili
  • 8% HuggingFaceH4/no_robots
  • 5% adversarial from kuzaai/kuza_sft_adversarial
  • all multiturn from kuzaai/kuza_sft_multiturn

LoRA: RsLoRA r=32, alpha=64, no QAT. Sequence length 1024, 2 epochs, LR 0.0002. Thinking is off (--reasoning off when llama.cpp supports it). GGUFs are text-only (MTP/nextn kept; vision/audio dropped). ssm_out is Q6_K on every candidate.

Files

  • adapter/ — PEFT adapter, tokenizer, SFT metrics and manifests
  • training/ — Trainer checkpoints including checkpoint-best
  • merged_bf16/ — text-only merged Hugging Face BF16 weights
  • reference/ — text-only BF16 GGUF (kuza-bf16.gguf) and smoke log
  • imatrix/ — calibration corpus, eval corpus, imatrix, logs
  • quants/ — quantized GGUF candidates:
  • q4_k_m_imatrix/kuza-qwen-q4_k_m.gguf (q4_k_m)
  • q4_k_xl_ssm/kuza-qwen-q4_k_xl.gguf (q4_k_m)
  • q4_k_s_ssm/kuza-qwen-q4_k_s.gguf (q4_k_s)
  • screen/ — GPU KLD diagnostic, hidden-set scores, and results.json winner
  • provenance/ — copied adapter metrics, recipes, screen JSON
  • upload_manifest.json — path, size, and sha256 for every uploaded file

Download

huggingface-cli download kuzaai/kuza-qwen-3.5-4b --local-dir ./kuza-qwen-3.5-4b

screen/results.json ranks by hidden-set accuracy, then GGUF size, then GPU TPS. Do not treat GPU TPS as an ADTC laptop measurement.

Identity and Version

Repository
kuzaai/kuza-qwen-3.5-4b
Publisher
Kuza AI
Task
Not stated by the source
Modality
Other
Library
peft
Parameters
Not stated by the source
Languages
en, sw
Revision
9c2d90a8f6a34df7e1bfe40eb85e5e655faa5284
First published
2026-09-18
Last updated
2026-09-18

Files and Weights

121 files, 33.2 GB in total. The weights are 33 files totalling 29.0 GB in bin, gguf, pt, pth, safetensors.

Weights33 files · 29.0 GB
Configuration38 files · 237.9 KB
Tokenizer14 files · 140.0 MB
Documentation8 files · 34.8 KB
Other27 files · 4.1 GB
Repository1 file · 2.4 KB
Every file
FileTypeSizeSHA-256
adapter/adapter_model.safetensorsWeights223.4 MB acb5d01c2f9a
merged_bf16/model-00001-of-00002.safetensorsWeights5.0 GB 8086e3be3757
merged_bf16/model-00002-of-00002.safetensorsWeights3.4 GB e1c20c3d73a0
provenance/adapter_model.safetensorsWeights223.4 MB acb5d01c2f9a
quants/q4_k_m_imatrix/kuza-qwen-q4_k_m.ggufWeights2.7 GB 4a7a8921cb83
quants/q4_k_s_ssm/kuza-qwen-q4_k_s.ggufWeights2.6 GB a8b7cfe78182
quants/q4_k_xl_ssm/kuza-qwen-q4_k_xl.ggufWeights3.0 GB 82fa31b2fa8c
reference/kuza-bf16.ggufWeights8.4 GB 8d355227aea4
training/checkpoint-2180/adapter_model.safetensorsWeights223.4 MB 117561a10de2
training/checkpoint-2180/optimizer.ptWeights447.0 MB 92de09b702f1
training/checkpoint-2180/rng_state.pthWeights14.6 KB 52e8d3009833
training/checkpoint-2180/scheduler.ptWeights1.5 KB 774c899a3c80
training/checkpoint-2180/training_args.binWeights5.8 KB f102db021651
training/checkpoint-4360/adapter_model.safetensorsWeights223.4 MB fdd033b851ca
training/checkpoint-4360/optimizer.ptWeights447.0 MB 3e53422a50d3
training/checkpoint-4360/rng_state.pthWeights14.6 KB c72427f97e81
training/checkpoint-4360/scheduler.ptWeights1.5 KB 54dc364ea571
training/checkpoint-4360/training_args.binWeights5.8 KB f102db021651
training/checkpoint-6540/adapter_model.safetensorsWeights223.4 MB acb5d01c2f9a
training/checkpoint-6540/optimizer.ptWeights447.0 MB d8b0a7c52f96
training/checkpoint-6540/rng_state.pthWeights14.6 KB 7ecff8432e52
training/checkpoint-6540/scheduler.ptWeights1.5 KB 585326863b1a
training/checkpoint-6540/training_args.binWeights5.8 KB f102db021651
training/checkpoint-8718/adapter_model.safetensorsWeights223.4 MB 7e404dcac763
training/checkpoint-8718/optimizer.ptWeights447.0 MB 61b92b1f2c11
training/checkpoint-8718/rng_state.pthWeights14.6 KB ca0af8fb33f5
training/checkpoint-8718/scheduler.ptWeights1.5 KB bc8efb8ff5c3
training/checkpoint-8718/training_args.binWeights5.8 KB f102db021651
training/checkpoint-best/adapter_model.safetensorsWeights223.4 MB acb5d01c2f9a
training/checkpoint-best/optimizer.ptWeights447.0 MB d8b0a7c52f96
training/checkpoint-best/rng_state.pthWeights14.6 KB 7ecff8432e52
training/checkpoint-best/scheduler.ptWeights1.5 KB 585326863b1a
training/checkpoint-best/training_args.binWeights5.8 KB f102db021651
adapter/adapter_config.jsonConfiguration11.2 KB
adapter/data_quality.jsonConfiguration1.0 KB
adapter/kuza_system_prompt.jsonConfiguration883 B
adapter/sft_manifest.jsonConfiguration10.7 KB
adapter/train_metrics.jsonConfiguration198 B
adapter/trainer_state.jsonConfiguration8.8 KB
imatrix/corpus_manifest.jsonConfiguration495 B
merged_bf16/config.jsonConfiguration3.0 KB
merged_bf16/generation_config.jsonConfiguration141 B
merged_bf16/kuza_system_prompt.jsonConfiguration883 B
merged_bf16/model.safetensors.index.jsonConfiguration54.8 KB
provenance/adapter_config.jsonConfiguration11.2 KB
provenance/data_quality.jsonConfiguration1.0 KB
provenance/kuza_system_prompt.jsonConfiguration883 B
provenance/provenance_index.jsonConfiguration177 B
provenance/q4_k_m_imatrix_recipe.jsonConfiguration672 B
provenance/q4_k_s_ssm_recipe.jsonConfiguration683 B
provenance/q4_k_xl_ssm_recipe.jsonConfiguration706 B
provenance/sft_manifest.jsonConfiguration10.7 KB
provenance/train_metrics.jsonConfiguration198 B
provenance/trainer_state.jsonConfiguration8.8 KB
quants/q4_k_m_imatrix/recipe.jsonConfiguration672 B
quants/q4_k_s_ssm/recipe.jsonConfiguration683 B
quants/q4_k_xl_ssm/recipe.jsonConfiguration706 B
reference/kuza_system_prompt.jsonConfiguration883 B
reference/reference_manifest.jsonConfiguration668 B
screen/mtp_probe.jsonConfiguration324 B
training/checkpoint-2180/adapter_config.jsonConfiguration11.2 KB
training/checkpoint-2180/trainer_state.jsonConfiguration3.0 KB
training/checkpoint-4360/adapter_config.jsonConfiguration11.2 KB
training/checkpoint-4360/trainer_state.jsonConfiguration4.9 KB
training/checkpoint-6540/adapter_config.jsonConfiguration11.2 KB
training/checkpoint-6540/trainer_state.jsonConfiguration6.8 KB
training/checkpoint-8718/adapter_config.jsonConfiguration11.2 KB
training/checkpoint-8718/trainer_state.jsonConfiguration8.6 KB
training/checkpoint-best/adapter_config.jsonConfiguration11.2 KB
training/checkpoint-best/trainer_state.jsonConfiguration6.8 KB
upload_manifest.jsonConfiguration20.4 KB
README.mdDocumentation2.1 KB
adapter/README.mdDocumentation5.2 KB
training/README.mdDocumentation1.4 KB
training/checkpoint-2180/README.mdDocumentation5.2 KB
training/checkpoint-4360/README.mdDocumentation5.2 KB
training/checkpoint-6540/README.mdDocumentation5.2 KB
training/checkpoint-8718/README.mdDocumentation5.2 KB
training/checkpoint-best/README.mdDocumentation5.2 KB
adapter/chat_template.jinjaOther9.0 KB
adapter/heldout.jsonlOther6.5 MB
imatrix/calibration.txtOther1.9 MB
imatrix/eval.txtOther397.0 KB
imatrix/imatrix.logOther1.2 KB
imatrix/kuza.imatrixOther3.6 MB 87adb974a66d
merged_bf16/chat_template.jinjaOther9.0 KB
quants/q4_k_m_imatrix-smoke.logOther
quants/q4_k_m_imatrix.logOther67.9 KB
quants/q4_k_s_ssm-smoke.logOther
quants/q4_k_s_ssm.logOther67.9 KB
quants/q4_k_xl_ssm-smoke.logOther
quants/q4_k_xl_ssm.logOther80.8 KB
reference/smoke.logOther
screen/bf16-kld.logOther999 B
screen/kuza-bf16-reference.kldOther4.1 GB 901bc84b9f91
screen/q4_k_m_imatrix-bench.logOther790 B
screen/q4_k_m_imatrix-english_agriculture.logOther
screen/q4_k_m_imatrix-english_safety.logOther
screen/q4_k_m_imatrix-kld.logOther3.9 KB
screen/q4_k_m_imatrix-swahili_agriculture.logOther
screen/q4_k_m_imatrix-swahili_safety.logOther
training/checkpoint-2180/chat_template.jinjaOther9.0 KB
training/checkpoint-4360/chat_template.jinjaOther9.0 KB
training/checkpoint-6540/chat_template.jinjaOther9.0 KB
training/checkpoint-8718/chat_template.jinjaOther9.0 KB
training/checkpoint-best/chat_template.jinjaOther9.0 KB
.gitattributesRepository2.4 KB
adapter/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
adapter/tokenizer_config.jsonTokenizer7.2 KB
merged_bf16/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
merged_bf16/tokenizer_config.jsonTokenizer1.2 KB
training/checkpoint-2180/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
training/checkpoint-2180/tokenizer_config.jsonTokenizer7.2 KB
training/checkpoint-4360/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
training/checkpoint-4360/tokenizer_config.jsonTokenizer7.2 KB
training/checkpoint-6540/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
training/checkpoint-6540/tokenizer_config.jsonTokenizer7.2 KB
training/checkpoint-8718/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
training/checkpoint-8718/tokenizer_config.jsonTokenizer7.2 KB
training/checkpoint-best/tokenizer.jsonTokenizer20.0 MB 87a7830d63fc
training/checkpoint-best/tokenizer_config.jsonTokenizer7.2 KB

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
29.0 GB
Download from Kuza AI

Released by Kuza AI through its official repository on Hugging Face. Read the license.

Built From

  • Adapter of unsloth/Qwen3.5-4B
  • Derived from unsloth/Qwen3.5-4B

Memory Requirements

PrecisionWeights in memory
As published29.0 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About kuza-qwen-3.5-4b

Can I use kuza-qwen-3.5-4b commercially?

Yes. kuza-qwen-3.5-4b is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.