Recommended Models

July 31, 2026 · View on GitHub

Models are loaded from the Hugging Face Hub by default. The model key accepts a hub repo id. remote_filename is a glob that selects the quantization variant. Local .gguf paths work too.

Chat / Summarizer / Dialog / Lang-detect / Translate

These wrappers use generative chat models. Any instruction-tuned GGUF model works. Smaller models run faster on CPU.

Tiny (CI / low-RAM, < 200 MB)

Modelmodelremote_filenameNotes
Smol-Llama 101Mafrideva/Smol-Llama-101M-Chat-v1-GGUF*q2_k.gguf~45 MB, used in CI e2e tests
Lite-Mistral 150MOuteAI/Lite-Mistral-150M-v2-Instruct-GGUF*Q4_K_M.gguf~100 MB

Small (1 to 2 GB, good for Raspberry Pi 5)

Modelmodelremote_filenameNotes
Qwen2 0.5B InstructQwen/Qwen2-0.5B-Instruct-GGUF*q8_0.ggufmultilingual
TinyLlama 1.1BTheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF*Q4_K_M.ggufEnglish

Medium (4 to 8 GB)

Modelmodelremote_filenameNotes
Mistral 7B Instruct v0.3MaziyarPanahi/Mistral-7B-Instruct-v0.3-GGUF*Q4_K_M.ggufstrong general
Phi-3 Mini 4Kmicrosoft/Phi-3-mini-4k-instruct-gguf*Q4_K_M.gguf3.8B, very capable
Qwen2 7B InstructQwen/Qwen2-7B-Instruct-GGUF*Q4_K_M.ggufmultilingual

Translation-specific

Modelmodelremote_filenameNotes
TowerInstruct 7BTheBloke/TowerInstruct-7B-v0.1-GGUF*Q4_K_M.gguffine-tuned for translation

Portuguese

Modelmodelremote_filename
Gervásio 7B PTPTRichardErkhov/PORTULAN_-_gervasio-7b-portuguese-ptpt-decoder-gguf*Q4_K_M.gguf
CabraLlama3 8Bmradermacher/CabraLlama3-8b-GGUF*Q4_K_M.gguf
Bode 7B PT-BRrecogna-nlp/bode-7b-alpaca-pt-br-gguf*Q4_K_M.gguf

Catalan

Modelmodelremote_filename
CataLlama v0.2catallama/CataLlama-v0.2-Instruct-SFT-DPO-Merged-GGUF*Q4_K_M.gguf

Embeddings

GGUFEmbeddings supports friendly names from GGUFEmbeddings.DEFAULT_MODELS:

Tiny (< 30 MB)

Friendly nameRepoDimsNotes
all-MiniLM-L6-v2leliuga/all-MiniLM-L6-v2-GGUF384used in CI e2e tests
e5-small-v2ChristianAzinn/e5-small-v2-gguf384
gte-smallChristianAzinn/gte-small-gguf384

Small (30 to 100 MB)

Friendly nameRepoDimsNotes
all-MiniLM-L12-v2leliuga/all-MiniLM-L12-v2-GGUF384
bge-small-en-v1.5ChristianAzinn/bge-small-en-v1.5-gguf384English
gte-baseChristianAzinn/gte-base-gguf768
snowflake-arctic-embed-xsChristianAzinn/snowflake-arctic-embed-xs-gguf384
snowflake-arctic-embed-sChristianAzinn/snowflake-arctic-embed-s-gguf384

Multilingual

Friendly nameRepoDimsNotes
labseChristianAzinn/labse-gguf768109 languages, default
nomic-embed-text-v1.5nomic-ai/nomic-embed-text-v1.5-GGUF768

Large / high-quality

Friendly nameRepoDimsNotes
bge-large-en-v1.5ChristianAzinn/bge-large-en-v1.5-gguf1024English
mxbai-embed-large-v1ChristianAzinn/mxbai-embed-large-v1-gguf1024
uae-large-v1ChristianAzinn/uae-large-v1-gguf1024
gte-largeChristianAzinn/gte-large-gguf1024
gte-Qwen2-1.5B-instructsecond-state/gte-Qwen2-1.5B-instruct-GGUF1536multilingual

These are community-maintained GGUF quantizations. Check the linked repos for licensing details before use in production.


← Localized prompts · Home