← All models

Talkie 1930 (13B)

generalApache 2.0reco 5

A 13B instruct model trained solely on pre-1931 English-language text — it writes and converses in an authentic period voice, free of modern anachronisms. Best for creative writing, correspondence, and in-character chat rather than tool use or current-events questions. Runs a community GGUF that rescales talkie's out-of-range layer-14 ffn_down channel so K-quantization stays coherent. 2K context. Apache-2.0.

At a glance

Parameters
13B
Quantization
Q4_K_M
Context window
2K tokens
Approx. size
8.6 GB
Engines
llama.cpp
License
Apache-2.0 (open)
Version
v1.0.0 , released 2026-07-20
Family
other

Tuning

Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.

sampling.temperature
0.8
sampling.topP
0.95
sampling.topK
40
sampling.maxTokens
1024
sampling.repetitionPenalty
1.1
sampling.repetitionContext
64
reasoning.enableThinking
false
profiles
creative

Model behaviors

Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).

Sources

llama.cpp (GGUF)
zakarth/talkie-1930-13b-it-vulkan-fixed-GGUF · talkie-1930-13b-it.Q4_K_M.gguf · Q4_K_M
Upstream
https://huggingface.co/talkie-lm/talkie-1930-13b-it
Manifest
View on GitHub

Tags