Skip to page content
SAVRN
Search Plan a project

Open-weight model · Text generation

qwen1.5b-int4-Octuma

by Omar Imanol Rodríguez Rodríguez Imanol11/qwen1.5b-int4-Octuma

qwen1.5b-int4-Octuma is an open-weight model for text generation from Omar Imanol Rodríguez Rodríguez, released under Apache License 2.0. It has 32,768-token context. Its published files total 2.6 GB. It draws 21 downloads a month.

Built with Qwen. Qwen/Qwen2.5-1.5B-Instruct cuantizado a INT4 con (2.34x mas chico) perdiendo 2.11% de calidad. Incluye las dos formas de usarlo: qwen1.5b-int4.gguf para llama.cpp y Android, y la carpeta.tq para PyTorch.

Parameters—
Context32,768
Weights2.6 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads21

Model Card

By Omar Imanol Rodríguez Rodríguez, published under apache-2.0, revision 50e712a29d20.

Built with Qwen. Qwen/Qwen2.5-1.5B-Instruct cuantizado a INT4 con (2.34x mas chico) perdiendo 2.11% de calidad. Incluye las dos formas de usarlo: qwen1.5b-int4.gguf para llama.cpp y Android, y la carpeta.tq para PyTorch. Perplejidad en wikitext-2 test, 20 ventanas de 2048 tokens, todo medido dentro de llama.cpp con llama-perplexity para que la comparacion sea de tu a tu. Este modelo pierde 2.2 veces menos calidad que llama.cpp Q4KM, el formato mas usado para correr modelos en local, a cambio de 0.34 GB mas de archivo. La comparativa de los tres tamaños esta en En PyTorch el modelo cuantizado genera mas lento que el original: Octuma desempaqueta los 4 bits en cada multiplicacion sin un…

Read Omar Imanol Rodríguez Rodríguez's full model card

Built with Qwen.

Qwen/Qwen2.5-1.5B-Instruct cuantizado a INT4 con Octuma: 3.09 GB → 1.32 GB (2.34x mas chico) perdiendo 2.11% de calidad.

Incluye las dos formas de usarlo: qwen1.5b-int4.gguf para llama.cpp y Android, y la carpeta .tq para PyTorch.

Calidad medida

Perplejidad en wikitext-2 test, 20 ventanas de 2048 tokens, todo medido dentro de llama.cpp con llama-perplexity para que la comparacion sea de tu a tu.

Formato Perplejidad Calidad perdida Tamaño
F16 (original) 8.3111 — 3.09 GB
Octuma INT4 (este modelo) 8.4864 +2.11% 1.32 GB
llama.cpp Q4_K_M 8.7047 +4.74% 0.99 GB
llama.cpp Q4_0 9.0051 +8.35% 0.93 GB

Este modelo pierde 2.2 veces menos calidad que llama.cpp Q4_K_M, el formato mas usado para correr modelos en local, a cambio de 0.34 GB mas de archivo.

Las perplejidades solo se comparan entre si cuando salen del mismo motor y el mismo corpus. Cada motor trocea y promedia distinto, asi que no contrastes estos numeros con los de otra ficha sin revisar como se midieron.

La comparativa de los tres tamaños esta en runs/COMPARATIVA_GGUF.md.

Como usarlo

llama.cpp (tambien en Android)

hf download Imanol11/qwen1.5b-int4-Octuma --local-dir qwen1.5b
llama-cli -m qwen1.5b/qwen1.5b-int4.gguf -p "Hola"

PyTorch

pip install "octuma[hf]"
from huggingface_hub import snapshot_download
from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer

from octuma.export.tq import load_quantized

ruta = snapshot_download("Imanol11/qwen1.5b-int4-Octuma")
cfg = AutoConfig.from_pretrained(ruta)
modelo = load_quantized(AutoModelForCausalLM.from_config(cfg), ruta)
tok = AutoTokenizer.from_pretrained(ruta)

En PyTorch el modelo cuantizado genera mas lento que el original: Octuma desempaqueta los 4 bits en cada multiplicacion sin un kernel dedicado. Para velocidad, usa el .gguf con llama.cpp.

Como se hizo

octuma quantize Qwen/Qwen2.5-1.5B-Instruct
octuma export <carpeta> --out qwen1.5b-int4.gguf
python scripts/verify_gguf.py <carpeta> qwen1.5b-int4.gguf

GPTQ + AWQ con grupos de 32 pesos, calibrado con wikitext-2/train. Octuma cuantiza por grupos con punto cero, reparte el error de redondeo entre las columnas pendientes (GPTQ) y escala los canales segun su importancia en las activaciones (AWQ).

Historial

El .gguf publicado aqui antes del 2026-09-20 estaba degradado: tenia tres metadatos mal escritos (rope.freq_base, el pre-tokenizador y el eos_token_id). Los pesos siempre estuvieron bien; el archivo actual los reescribe correctamente. Si lo descargaste antes de esa fecha, vuelve a bajarlo.

El 2026-10-07 se re-exporto con Octuma 0.1.5, sin recuantizar: los pesos y la perplejidad son los mismos. Cambian tres metadatos: el archivo ya trae la plantilla de chat (tokenizer.chat_template), se anuncia como Q4_1 en vez de F16, y los metadatos de la carpeta .tq pasan de tinyq.json a octuma.json. El .gguf pesa 2 560 bytes mas y su SHA-256 es otro.

Licencia

Los pesos derivan de Qwen/Qwen2.5-1.5B-Instruct, bajo Apache 2.0; la copia del original va en LICENSE. El codigo de Octuma es MIT.

Los archivos de pesos estan modificados respecto al original: cuantizados a INT4 con Octuma.

Configuration

Architecture
Qwen2ForCausalLM
Context length (tokens)
32,768
Layers
28
Hidden size
1,536
Feed-forward size
8,960
Attention heads
12
Key/value heads
2
Vocabulary size
151,936
Model type
qwen2

Identity and Version

Repository
Imanol11/qwen1.5b-int4-Octuma
Publisher
Omar Imanol Rodríguez Rodríguez
Task
Text generation
Modality
Text
Library
gguf
Parameters
Not stated by the source
Languages
en, es
Revision
50e712a29d20688146c1f1bd80a6aac8a47870c9
First published
2026-09-20
Last updated
2026-10-08

Files and Weights

10 files, 2.6 GB in total. The weights are 2 files totalling 2.6 GB in gguf, safetensors.

Weights2 files · 2.6 GB
Configuration2 files · 33.7 KB
Tokenizer2 files · 11.4 MB
Documentation2 files · 15.3 KB
Other1 file · 2.5 KB
Repository1 file · 1.6 KB
Every file
FileTypeSizeSHA-256
model.tq.safetensorsWeights1.2 GB 2bd7f84679bf
qwen1.5b-int4.ggufWeights1.3 GB 696021abd23f
config.jsonConfiguration1.4 KB —
octuma.jsonConfiguration32.4 KB —
LICENSEDocumentation11.3 KB —
README.mdDocumentation3.9 KB —
chat_template.jinjaOther2.5 KB —
.gitattributesRepository1.6 KB —
tokenizer.jsonTokenizer11.4 MB 3fd169731d2c
tokenizer_config.jsonTokenizer694 B —

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
2.6 GB
Download from Omar Imanol Rodríguez Rodríguez

Released by Omar Imanol Rodríguez Rodríguez through its official repository on Hugging Face. Read the license.

Built From

Memory Requirements

PrecisionWeights in memory
As published2.6 GB

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

Questions About qwen1.5b-int4-Octuma

Can I use qwen1.5b-int4-Octuma commercially?

Yes. qwen1.5b-int4-Octuma 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.

What is qwen1.5b-int4-Octuma's context length?

32,768 tokens, from the maximum position embeddings in its published configuration.

Similar Models

Model · Text generation

opt-125m

AI at Meta

OPT was first introduced in Open Pre-trained Transformer Language Models and first released in metaseq's repository on May 3rd 2022 by Meta AI. Disclaimer: The team releasing OPT wrote an official model card, which is available in Appendix D of the paper. Content from this model card has been written by the Hugging Face team. To quote the first two paragraphs of the official paper OPT was predominantly pretrained with English text, but a small amount of non-English data is still present within the training corpus via CommonCrawl. The model was pretrained using a causal language modeling (CLM) objective. OPT belongs to the same family of decoder-only models like GPT-3. As such, it was…

Open weights other 2,048 tokens transformers

Model · Text generation

Ornith-1.5-9B-GGUF

Ornith

Chirp Chirp! We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement. Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For…

Open weights mit transformers

Model · Text generation

Ornith-1.5-35B-A3B-GGUF

Ornith

Chirp Chirp! We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement. Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For…

Open weights mit transformers

Model · Text generation

Ternary-Bonsai-2-27B-gguf

Prism ML

Full 27B-class reasoning in ternary transformer weights, for llama.cpp (CUDA, Metal, CPU) - \~5.9 GB language model (down from \~54 GB FP16) — full 27B-class reasoning on a standard laptop or a single GPU - 98.2% of FP16 intelligence retained: 84.78 average across 14 thinking-mode benchmarks — far above the conventional IQ2XXS build (72.59) at about 82% of its footprint, and within 0.4 points of UD-Q4KXL at three times the footprint - Retains thinking, reasoning, and agentic behavior deep in the sub-4-bit regime, where conventional low-bit representations collapse: math within half a point of full precision (96.57), coding level with the baseline (89.42), agentic tool calling at 74.92…

Open weights apache-2.0 llama.cpp

Uncensored Qwen3.8-27B, published as GGUF quantizations with the multi token prediction (MTP) head retained and verified. Refusal behaviour has been substantially reduced, not eliminated. See Measured behaviour for the numbers. Capabilities, training data, and architecture are otherwise unchanged. - Refusal directions removed with Heretic, which co minimizes refusal count against KL divergence from the base model. No handwritten refusal removal code, no finetuning, no additional training data. - Abliteration runs at bf16 (no 4 bit quantization). the resulting LoRA is merged into the bf16 base, so the published weights are not a quantized round trip. - mtp. tensors are copied verbatim from…

Open weights apache-2.0 llama.cpp

Model · Text generation

Ornith-1.0-9B-GGUF

Ornith

Aloha! Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding. This model card documents Ornith-1.0-9B, the most lightweight member of the Ornith family, designed for efficient single-GPU deployment. Ornith-1.0-9B is a dense ~9B model (≈19 GB in bf16), so it serves comfortably on a single 80GB GPU. The recipes below stand up an OpenAI-compatible server; add --tensor-parallel-size / --tp if you want to shard across more GPUs. For a quick local test (or to script offline generation), load the model directly with Transformers. Make sure you have a recent release installed — see the Transformers installation guide; Ornith-1.0-9B requires…

Open weights mit transformers