Instructions to use sandeeprdy1729/thrisul-parser with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sandeeprdy1729/thrisul-parser with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="sandeeprdy1729/thrisul-parser", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("sandeeprdy1729/thrisul-parser", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use sandeeprdy1729/thrisul-parser with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "sandeeprdy1729/thrisul-parser"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sandeeprdy1729/thrisul-parser",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/sandeeprdy1729/thrisul-parser

SGLang

How to use sandeeprdy1729/thrisul-parser with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "sandeeprdy1729/thrisul-parser" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sandeeprdy1729/thrisul-parser",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "sandeeprdy1729/thrisul-parser" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sandeeprdy1729/thrisul-parser",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use sandeeprdy1729/thrisul-parser with Docker Model Runner:
```
docker model run hf.co/sandeeprdy1729/thrisul-parser
```

thrisul-parser / nemotron_parse /engine /header_inference_probe.py

sandeeprdy1729

Thrisul-Parser v0.1

37fb2b9 verified 20 days ago

Raw History Blame Contribute Delete

3.38 kB

# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: OpenMDW-1.1

"""Infer leading header rows from predicted table spans.
No benchmark labels, source-image metadata, or dataset IDs are accepted as input.
"""
import re
from bs4 import BeautifulSoup


def infer_header_rows(raw):
    soup = BeautifulSoup(raw, "html.parser")
    diagnostics = []
    for table in soup.find_all("table"):
        if table.find_parent("table"):
            continue
        rows = [r for r in table.find_all("tr") if r.find_parent("table") == table]
        if not rows:
            continue
        # Respect existing explicit cell roles; this probe is only for td-only output.
        if table.find("th"):
            diagnostics.append({"changed": False, "reason": "existing_th"})
            continue
        start = 0
        while start < min(3, len(rows) - 1):
            cells = rows[start].find_all(["td", "th"], recursive=False)
            if (
                len(cells) == 1
                and int(cells[0].get("colspan", 1)) > 1
                and int(cells[0].get("rowspan", 1)) == 1
            ):
                start += 1
            else:
                break
        cells = rows[start].find_all(["td", "th"], recursive=False)
        if not cells:
            continue
        numeric = lambda s: bool(re.fullmatch(r"[\s\d.,%$€£()+−-]+", s))
        texts = [
            c.get_text(" ", strip=True) for c in cells if c.get_text(" ", strip=True)
        ]
        if not texts or sum(numeric(t) for t in texts) / len(texts) > 0.5:
            diagnostics.append({"changed": False, "reason": "numeric_first_row"})
            continue
        end = start + 1
        rs = max(int(c.get("rowspan", 1)) for c in cells)
        cs = max(int(c.get("colspan", 1)) for c in cells)
        if 1 < rs <= 4:
            end = min(len(rows), start + rs)
        elif rs > 4:
            diagnostics.append({"changed": False, "reason": "long_rowspan"})
            continue
        elif cs > 1 and start + 1 < len(rows):
            nxt = [
                c
                for c in rows[start + 1].find_all(["td", "th"], recursive=False)
                if c.get_text(" ", strip=True)
            ]
            # A group heading's colspan does not establish the next row's role.
            # Require explicit header styling on every nonempty next-row cell;
            # text-heavy data rows and year-valued data must not become headers.
            if (
                len(nxt) >= 2
                and all(c.find(["b", "strong"]) for c in nxt)
                and sum(numeric(c.get_text(" ", strip=True)) for c in nxt) / len(nxt)
                < 0.5
            ):
                end = start + 2
        # Only probe multi-level headers. Single-row tables remain byte-for-byte as supplied.
        if end - start <= 1:
            diagnostics.append({"changed": False, "reason": "single_level"})
            continue
        for row in rows[start:end]:
            for c in row.find_all(["td", "th"], recursive=False):
                c.name = "th"
        diagnostics.append(
            {
                "changed": True,
                "start": start,
                "end": end,
                "rule": "bounded_span_header_topology",
            }
        )
    return str(soup), diagnostics

	# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
	# SPDX-License-Identifier: OpenMDW-1.1

	"""Infer leading header rows from predicted table spans.
	No benchmark labels, source-image metadata, or dataset IDs are accepted as input.
	"""
	import re
	from bs4 import BeautifulSoup


	def infer_header_rows(raw):
	soup = BeautifulSoup(raw, "html.parser")
	diagnostics = []
	for table in soup.find_all("table"):
	if table.find_parent("table"):
	continue
	rows = [r for r in table.find_all("tr") if r.find_parent("table") == table]
	if not rows:
	continue
	# Respect existing explicit cell roles; this probe is only for td-only output.
	if table.find("th"):
	diagnostics.append({"changed": False, "reason": "existing_th"})
	continue
	start = 0
	while start < min(3, len(rows) - 1):
	cells = rows[start].find_all(["td", "th"], recursive=False)
	if (
	len(cells) == 1
	and int(cells[0].get("colspan", 1)) > 1
	and int(cells[0].get("rowspan", 1)) == 1
	):
	start += 1
	else:
	break
	cells = rows[start].find_all(["td", "th"], recursive=False)
	if not cells:
	continue
	numeric = lambda s: bool(re.fullmatch(r"[\s\d.,%$€£()+−-]+", s))
	texts = [
	c.get_text(" ", strip=True) for c in cells if c.get_text(" ", strip=True)
	]
	if not texts or sum(numeric(t) for t in texts) / len(texts) > 0.5:
	diagnostics.append({"changed": False, "reason": "numeric_first_row"})
	continue
	end = start + 1
	rs = max(int(c.get("rowspan", 1)) for c in cells)
	cs = max(int(c.get("colspan", 1)) for c in cells)
	if 1 < rs <= 4:
	end = min(len(rows), start + rs)
	elif rs > 4:
	diagnostics.append({"changed": False, "reason": "long_rowspan"})
	continue
	elif cs > 1 and start + 1 < len(rows):
	nxt = [
	c
	for c in rows[start + 1].find_all(["td", "th"], recursive=False)
	if c.get_text(" ", strip=True)
	]
	# A group heading's colspan does not establish the next row's role.
	# Require explicit header styling on every nonempty next-row cell;
	# text-heavy data rows and year-valued data must not become headers.
	if (
	len(nxt) >= 2
	and all(c.find(["b", "strong"]) for c in nxt)
	and sum(numeric(c.get_text(" ", strip=True)) for c in nxt) / len(nxt)
	< 0.5
	):
	end = start + 2
	# Only probe multi-level headers. Single-row tables remain byte-for-byte as supplied.
	if end - start <= 1:
	diagnostics.append({"changed": False, "reason": "single_level"})
	continue
	for row in rows[start:end]:
	for c in row.find_all(["td", "th"], recursive=False):
	c.name = "th"
	diagnostics.append(
	{
	"changed": True,
	"start": start,
	"end": end,
	"rule": "bounded_span_header_topology",
	}
	)
	return str(soup), diagnostics