dlasher/humanizer-ENG:IQ4

dlasher/ humanizer-ENG:IQ4_NL

55 Downloads Updated 1 month ago

pulled as BF16, vocab slice (text_en_large) requant to IQ4_NL (text_en_large) -- (English only focus) - roughly ~15% faster than stock IQ4_NL

tools thinking

ollama run dlasher/humanizer-ENG:IQ4_NL

curl http://localhost:11434/api/chat \
  -d '{
    "model": "dlasher/humanizer-ENG:IQ4_NL",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='dlasher/humanizer-ENG:IQ4_NL',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'dlasher/humanizer-ENG:IQ4_NL',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 month ago

1 month ago

d8e6aab05fe5 · 6.4GB

model

archgemma4

parameters11.2B

quantizationIQ4_NL

6.4GB

params

{ "min_p": 0.05, "num_batch": 2048, "num_ctx": 262144, "presence_penalty": 0, "r

242B

template

{{- if .System }}<|im_start|>system {{ .System }}<|im_end|> {{- end }} {{- range .Messages }} {{- if

286B

Readme

PERPLEXITY QUALITY SCORECARD

Base Model PPL : 6.8998 (16-bit reference)

Quantized Model PPL : 7.0111 (IQ4_NL target)

Perplexity Drift : +0.1113 (+1.61%)

Quality Assessment : [PASS] EXCELLENT (Near-Lossless / Gold Standard)

NOTE: This model has been heavily vocab-sliced to reduce non-English tokens. YMMV

Original Vocabulary Size : 262,144 tokens

Pruned Vocabulary Size : 78,612 tokens

Reduction Ratio : 70.01% removed (183,532 tokens dropped)

BPE Merges Filtered : 151,542 / 514,906 rules retained

humanizer A 12B model that rewrites AI-written drafts (emails, essays, reports, forum posts; English and Chinese) so they read like a person wrote them. It is trained to keep every number, unit, date, name and quote and to add nothing, and it runs locally.

SOURCES:

** https://huggingface.co/jialinyyzz/humanizer

** https://huggingface.co/datasets/eaddario/imatrix-calibration

** https://github.com/sgaofen/humanizer-local-model/

`PERPLEXITY QUALITY SCORECARD`

` Base Model PPL      : 6.8998 (16-bit reference)`

` Quantized Model PPL : 7.0111 (IQ4_NL target)`

` Perplexity Drift    : +0.1113 (+1.61%)`

` Quality Assessment  : [PASS] EXCELLENT (Near-Lossless / Gold Standard)`

**NOTE: This model has been heavily vocab-sliced to reduce non-English tokens. YMMV**

`Original Vocabulary Size : 262,144 tokens`

`Pruned Vocabulary Size   : 78,612 tokens`

`Reduction Ratio          : 70.01% removed (183,532 tokens dropped)`

`BPE Merges Filtered      : 151,542 / 514,906 rules retained`

![image.png](/assets/dlasher/humanizer/aae4d3db-b4e9-41b9-b715-b1862d7a8b6d)

humanizer
A 12B model that rewrites AI-written drafts (emails, essays, reports, forum posts; English and Chinese) so they read like a person wrote them. It is trained to keep every number, unit, date, name and quote and to add nothing, and it runs locally.

![image.png](/assets/dlasher/humanizer/36f5127d-aa62-4c38-abe5-bd9ca9859f91)

SOURCES:

** https://huggingface.co/jialinyyzz/humanizer

** https://huggingface.co/datasets/eaddario/imatrix-calibration

** https://github.com/sgaofen/humanizer-local-model/

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)