Instructions to use Mike0021/pulpie-orange-small-gguf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Mike0021/pulpie-orange-small-gguf with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Mike0021/pulpie-orange-small-gguf",
	filename="pulpie-orange-small-F16.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Mike0021/pulpie-orange-small-gguf with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Mike0021/pulpie-orange-small-gguf:Q4_K_M

Use Docker

docker model run hf.co/Mike0021/pulpie-orange-small-gguf:Q4_K_M

LM Studio
Jan
Ollama
How to use Mike0021/pulpie-orange-small-gguf with Ollama:
```
ollama run hf.co/Mike0021/pulpie-orange-small-gguf:Q4_K_M
```

Unsloth Studio

How to use Mike0021/pulpie-orange-small-gguf with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Mike0021/pulpie-orange-small-gguf to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Mike0021/pulpie-orange-small-gguf to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Mike0021/pulpie-orange-small-gguf to start chatting

Atomic Chat new
Docker Model Runner
How to use Mike0021/pulpie-orange-small-gguf with Docker Model Runner:
```
docker model run hf.co/Mike0021/pulpie-orange-small-gguf:Q4_K_M
```

Lemonade

How to use Mike0021/pulpie-orange-small-gguf with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Mike0021/pulpie-orange-small-gguf:Q4_K_M

Run and chat with the model

lemonade run user.pulpie-orange-small-gguf-Q4_K_M

List all available models

lemonade list

Mike0021 commited on 1 day ago

Commit

403290b

verified ·

1 Parent(s): 78b487c

Clean up README formatting

Browse files

Files changed (1) hide show

README.md +38 -43

README.md CHANGED Viewed

@@ -11,82 +11,77 @@ tags:
 - pulpie
 ---
-# Pulpie Orange Small GGUF
-GGUF conversion of `feyninc/pulpie-orange-small`, a 210M parameter EuroBERT token-classification model for Pulpie HTML main-content extraction.
-This is an encoder classifier, not a causal language model. The GGUF files expose raw per-token `other/main` classifier logits through `llama-embedding --pooling none --embd-normalize -1`.
-## Files
-| File | Size MB | Avg GGUF subprocess latency s | Max abs diff vs PyTorch |
-|---|---:|---:|---:|
-| `pulpie-orange-small-F16.gguf` | 431.40 | 2.550 | 0.06980780208435067 |
-| `pulpie-orange-small-Q2_K.gguf` | 130.18 | 14.049 |  |
-| `pulpie-orange-small-Q3_K_M.gguf` | 143.15 | 11.036 |  |
-| `pulpie-orange-small-Q4_K_M.gguf` | 156.97 | 7.651 |  |
-| `pulpie-orange-small-Q5_K_M.gguf` | 168.91 | 2.640 |  |
-| `pulpie-orange-small-Q6_K.gguf` | 181.60 | 2.614 |  |
-| `pulpie-orange-small-Q8_0.gguf` | 232.89 | 2.630 | 0.0720623857818603 |
-PyTorch baseline latency samples: `[0.1122, 0.1968, 0.1134, 0.1765, 0.1171]` seconds.
 ## Verification
-- F16 max abs diff: `0.06980780208435067`
-- Q8_0 max abs diff: `0.0720623857818603`
-- F16 extraction non-empty: `True`, main blocks: `3`, markdown: `<html><body>
-    <main><h1 _item_id="1">Orange harvest report</h1>
-      <p _item_id="2">The main article explains how growers sort oranges after harvest.</p>
-      <p _item_id="3">It includes storage notes and quality checks for shipment.</p>
-    </main></body></html>`
-- Q8_0 extraction non-empty: `True`, main blocks: `3`, markdown: `<html><body>
-    <main><h1 _item_id="1">Orange harvest report</h1>
-      <p _item_id="2">The main article explains how growers sort oranges after harvest.</p>
-      <p _item_id="3">It includes storage notes and quality checks for shipment.</p>
-    </main></body></html>`
-Full machine-readable results are in `verification_report.json`.
-## llama.cpp Usage
 ```bash
 llama-embedding \
   --model pulpie-orange-small-Q8_0.gguf \
-  --prompt '<html><body><p _item_id="0">Main text.</p></body></html><|sep|>' \
   --pooling none \
   --embd-normalize -1 \
   --embd-output-format json
 ```
-Each returned embedding row is `[other_logit, main_logit]`. Pulpie classifies each block at the `<|sep|>` token position and keeps blocks where `main_logit > other_logit`.
-`llama-cli` can load the files, but it is a generation frontend and these GGUFs do not contain a causal LM head.
-## Python Sketch
 ```python
 import json, re, subprocess
-cmd = [
     "llama-embedding", "-m", "pulpie-orange-small-Q8_0.gguf",
     "-p", html_chunk,
     "--pooling", "none",
     "--embd-normalize", "-1",
     "--embd-output-format", "json",
-]
-out = subprocess.check_output(cmd, text=True)
-data = json.loads(re.search(r'\{\s*"object".*', out, re.S).group(0))
 logits = [row["embedding"] for row in data["data"]]
 ```
-## Conversion Method
-Current llama.cpp supports the EuroBERT encoder architecture but not the `EuroBertForTokenClassification` dispatch/head directly. The conversion used a small llama.cpp patch that:
-- registers `EuroBertForTokenClassification` in the HF converter,
-- maps `classifier.weight` and `classifier.bias` to GGUF `cls.output.*`,
-- writes classifier output labels `["other", "main"]` and `embedding_length_out=2`,
-- applies the classifier head in the EuroBERT runtime graph and returns per-token logits as embeddings.
-Quantization was performed with `llama-quantize` from the patched llama.cpp build.

 - pulpie
 ---
+# Pulpie Orange Small — GGUF
+GGUF conversion of [feyninc/pulpie-orange-small](https://huggingface.co/feyninc/pulpie-orange-small), a 210M parameter EuroBERT token-classification model for Pulpie HTML main-content extraction.
+## ⚠️ Not a language model
+This is an **encoder classifier**, not a causal LM. The GGUF files expose per-token `other`/`main` classifier logits via `llama-embedding`, not `llama-cli` generation.
+## Files
+| Quant | Size | Notes |
+|-------|------|-------|
+| F16 | 431 MB | Full precision baseline |
+| Q8_0 | 233 MB | 8-bit, verified accurate |
+| Q6_K | 182 MB | |
+| Q5_K_M | 169 MB | |
+| Q4_K_M | 157 MB | |
+| Q3_K_M | 143 MB | |
+| Q2_K | 130 MB | Most aggressive quantization |
 ## Verification
+| Variant | Max diff vs PyTorch | E2E extraction | Prediction agreement |
+|---------|--------------------|-----------------|---------------------|
+| F16 | 0.070 | ✅ 3 main blocks | 100% |
+| Q8_0 | 0.072 | ✅ 3 main blocks | 100% |
+| Q6_K – Q2_K | not tested | not tested | — |
+> **Note:** Only F16 and Q8_0 were numerically verified against the original PyTorch model. Lower quants (Q6_K → Q2_K) passed load + inference checks but output consistency was not validated. Use F16 or Q8_0 for production.
+Full results: [`verification_report.json`](./verification_report.json)
+## Usage
 ```bash
 llama-embedding \
   --model pulpie-orange-small-Q8_0.gguf \
+  --prompt "<html><body><p>Hello world</p></body></html><|sep|>" \
   --pooling none \
   --embd-normalize -1 \
   --embd-output-format json
 ```
+Each output row is `[other_logit, main_logit]`. Pulpie classifies at `<|sep|>` token positions and keeps blocks where `main_logit > other_logit`.
+`llama-cli` can load the files but cannot generate — these GGUFs have no causal LM head.
+## Python example
 ```python
 import json, re, subprocess
+out = subprocess.check_output([
     "llama-embedding", "-m", "pulpie-orange-small-Q8_0.gguf",
     "-p", html_chunk,
     "--pooling", "none",
     "--embd-normalize", "-1",
     "--embd-output-format", "json",
+], text=True)
+data = json.loads(re.search(r'\{.*\}', out, re.S).group(0))
 logits = [row["embedding"] for row in data["data"]]
 ```
+## Conversion notes
+Stock llama.cpp supports EuroBERT embeddings but not `EuroBertForTokenClassification`. A 103-line patch was applied to:
+- Register `EuroBertForTokenClassification` in the HF converter
+- Map `classifier.weight`/`classifier.bias` → GGUF `cls.output.*`
+- Write classifier labels `["other", "main"]` with `embedding_length_out=2`
+- Apply the classifier head in the EuroBERT runtime graph
+Quantization was done with `llama-quantize` from the patched build.