Datasets:

gesh75
/

multivendor-cli-commands

---
license: mit
pretty_name: Multivendor CLI Commands
language:
  - en
tags:
  - networking
  - network-automation
  - cli
  - cisco
  - juniper
  - arista
  - frr
size_categories:
  - 10K<n<100K
task_categories:
  - text-generation
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.parquet
---
# Multivendor CLI Commands

**70,006 network CLI commands across 17 vendors and tools, tagged with vendor, OS, device role and category.** The command corpus behind the [Multivendor CLI Configurator](https://huggingface.co/spaces/gesh75/multivendor-cli-configurator), as one flat Parquet table.

Source: [github.com/gesh75/multivendor-cli-configurator](https://github.com/gesh75/multivendor-cli-configurator) · Docs: [gesh75.github.io/multivendor-cli-configurator](https://gesh75.github.io/multivendor-cli-configurator/)

## Schema

| column | type | meaning |
|---|---|---|
| `vendor` | string | Vendor or tool (Cisco, Juniper, Arista, FRR, …) |
| `os` | string | Platform / OS surface (`ios`, `junos`, `eos`, `sros`, …) |
| `role` | string | `router`, `switch` or `firewall` |
| `category` | string | Functional bucket assigned by the pipeline (BGP, OSPF, VLAN, Interfaces, Security, …) |
| `command` | string | The command syntax, with placeholders (`[IP]`, `<vlan>`) or example values |
| `live_verified` | bool | `true` if captured from a live 10-node Docker FRR lab |
| `in_docs` | bool / null | `true` if the FRR command was also found in FRR docs; null elsewhere |

## Rows by vendor

| vendor | rows | vendor | rows |
|---|---:|---|---:|
| Cisco | 22,168 | Extreme | 2,800 |
| Arista | 7,397 | Linux | 2,592 |
| VyOS | 6,289 | FortiOS | 2,383 |
| Huawei | 4,425 | Wireshark | 2,130 |
| Aruba | 4,106 | PAN-OS | 1,224 |
| FRR | 3,949 | Mikrotik | 1,216 |
| Microsoft | 3,352 | NVIDIA | 1,168 |
| Juniper | 3,217 | Nokia | 1,131 |
| | | SONiC | 459 |

## Provenance — what was kept, what was dropped

The commands were collected from public vendor CLI references, a public command guide, community notes and a live FRR lab, then merged and deduped by a stdlib-only Python pipeline (`scripts/` in the repo).

This release keeps **only the command syntax and structural metadata**. The upstream `desc` (description prose) and `title` fields were **dropped**, because some of that text comes from or paraphrases vendor documentation and books, and its origin can't be traced row by row. If you need descriptions, use the vendor's own docs.

## License

MIT, matching the repo. CLI command names and syntax are functional facts belonging to their respective vendors; vendor and product names are trademarks of their owners. No affiliation or endorsement implied.

## Limitations

- Not validated per release or platform. Verify on your gear.
- Coverage is uneven: Cisco/Arista/VyOS/FRR are deep, SONiC and Nokia SR OS are thin.
- `category` comes from heuristics, so some rows are misbucketed (catch-all buckets like `Misc` and `Protocols` are large).
- Some commands carry a prompt or mode prefix (e.g. `switch(config-if)#`) inherited from the source.
- Example values are generic placeholders, documentation ranges or well-known public resolvers (e.g. `8.8.8.8`). No real network data.

## Derived

- [gesh75/netops-intent2cli](https://huggingface.co/datasets/gesh75/netops-intent2cli): synthetic operator intents paired with verbatim commands from this corpus. Default config is v1 (565 train / 102 validation / 108 test). Config `v2` adds 650 training pairs and an 88-row `operator_test`.
- [gesh75/netops-intent2cli-qwen0.5b-lora](https://huggingface.co/gesh75/netops-intent2cli-qwen0.5b-lora): experimental Qwen2.5-0.5B LoRA trained on v2. Adapter-only exact match: **12/108 (11.1%)** on the frozen v1 test, **9/88 (10.2%)** on `operator_test`. Optional retrieval+validator lookup mode (same card): **63/108** and **58/88** when the gold command is already in this corpus; it cannot return a command that is missing from the base. Not production-safe. Browser demo: [Intent finder](https://gesh75-multivendor-cli-configurator.static.hf.space/intent-finder.html) (retrieval only, no model).

## Load

```python
from datasets import load_dataset
ds = load_dataset("gesh75/multivendor-cli-commands", split="train")
bgp = ds.filter(lambda r: r["vendor"] == "Juniper" and r["category"] == "BGP")
```

```python
import pandas as pd
df = pd.read_parquet("hf://datasets/gesh75/multivendor-cli-commands/data/train.parquet")
```

---

### GESH · NetOps tools on Hugging Face

| | |
|---|---|
| 📡 **[netlog-ai](https://huggingface.co/spaces/gesh75/netlog-ai)** | Sanitize-first network log analysis — browser lite demo of [netlog-ai](https://pypi.org/project/netlog-ai/) (PyPI 2.1.0) |
| 🛠️ **[multivendor-cli-configurator](https://huggingface.co/spaces/gesh75/multivendor-cli-configurator)** | 70k+ CLI commands across 17 vendors — cheatsheet, N-vendor compare, CLI Studio, Intent finder |
| 🗂️ **[multivendor-cli-commands](https://huggingface.co/datasets/gesh75/multivendor-cli-commands)** | The same 70,006 commands as a Parquet dataset |
| 🧭 **[netops-intent2cli](https://huggingface.co/datasets/gesh75/netops-intent2cli)** | v1 + v2 synthetic operator intents → corpus command pairs |
| 🤖 **[netops-intent2cli-qwen0.5b-lora](https://huggingface.co/gesh75/netops-intent2cli-qwen0.5b-lora)** | LoRA adapter: English intent → vendor CLI (experimental) |

Collection: **[NetOps tools](https://huggingface.co/collections/gesh75/netops-tools-6aef9999054238474f968bcd)** · Code: [github.com/gesh75](https://github.com/gesh75) · Docs hub: [gesh75.github.io](https://gesh75.github.io/)

**Doctrine:** fail closed · sanitize first · evidence leaves, packets don't · honest coverage.

	---
	license: mit
	pretty_name: Multivendor CLI Commands
	language:
	- en
	tags:
	- networking
	- network-automation
	- cli
	- cisco
	- juniper
	- arista
	- frr
	size_categories:
	- 10K<n<100K
	task_categories:
	- text-generation
	configs:
	- config_name: default
	data_files:
	- split: train
	path: data/train.parquet
	---

	# Multivendor CLI Commands

	70,006 network CLI commands across 17 vendors and tools, tagged with vendor, OS, device role and category. The command corpus behind the [Multivendor CLI Configurator](https://huggingface.co/spaces/gesh75/multivendor-cli-configurator), as one flat Parquet table.

	Source: [github.com/gesh75/multivendor-cli-configurator](https://github.com/gesh75/multivendor-cli-configurator) · Docs: [gesh75.github.io/multivendor-cli-configurator](https://gesh75.github.io/multivendor-cli-configurator/)

	## Schema

	\| column \| type \| meaning \|
	\|---\|---\|---\|
	\| `vendor` \| string \| Vendor or tool (Cisco, Juniper, Arista, FRR, …) \|
	\| `os` \| string \| Platform / OS surface (`ios`, `junos`, `eos`, `sros`, …) \|
	\| `role` \| string \| `router`, `switch` or `firewall` \|
	\| `category` \| string \| Functional bucket assigned by the pipeline (BGP, OSPF, VLAN, Interfaces, Security, …) \|
	\| `command` \| string \| The command syntax, with placeholders (`[IP]`, `<vlan>`) or example values \|
	\| `live_verified` \| bool \| `true` if captured from a live 10-node Docker FRR lab \|
	\| `in_docs` \| bool / null \| `true` if the FRR command was also found in FRR docs; null elsewhere \|

	## Rows by vendor

	\| vendor \| rows \| vendor \| rows \|
	\|---\|---:\|---\|---:\|
	\| Cisco \| 22,168 \| Extreme \| 2,800 \|
	\| Arista \| 7,397 \| Linux \| 2,592 \|
	\| VyOS \| 6,289 \| FortiOS \| 2,383 \|
	\| Huawei \| 4,425 \| Wireshark \| 2,130 \|
	\| Aruba \| 4,106 \| PAN-OS \| 1,224 \|
	\| FRR \| 3,949 \| Mikrotik \| 1,216 \|
	\| Microsoft \| 3,352 \| NVIDIA \| 1,168 \|
	\| Juniper \| 3,217 \| Nokia \| 1,131 \|
	\| \| \| SONiC \| 459 \|

	## Provenance — what was kept, what was dropped

	The commands were collected from public vendor CLI references, a public command guide, community notes and a live FRR lab, then merged and deduped by a stdlib-only Python pipeline (`scripts/` in the repo).

	This release keeps only the command syntax and structural metadata. The upstream `desc` (description prose) and `title` fields were dropped, because some of that text comes from or paraphrases vendor documentation and books, and its origin can't be traced row by row. If you need descriptions, use the vendor's own docs.

	## License

	MIT, matching the repo. CLI command names and syntax are functional facts belonging to their respective vendors; vendor and product names are trademarks of their owners. No affiliation or endorsement implied.

	## Limitations

	- Not validated per release or platform. Verify on your gear.
	- Coverage is uneven: Cisco/Arista/VyOS/FRR are deep, SONiC and Nokia SR OS are thin.
	- `category` comes from heuristics, so some rows are misbucketed (catch-all buckets like `Misc` and `Protocols` are large).
	- Some commands carry a prompt or mode prefix (e.g. `switch(config-if)#`) inherited from the source.
	- Example values are generic placeholders, documentation ranges or well-known public resolvers (e.g. `8.8.8.8`). No real network data.

	## Derived

	- [gesh75/netops-intent2cli](https://huggingface.co/datasets/gesh75/netops-intent2cli): synthetic operator intents paired with verbatim commands from this corpus. Default config is v1 (565 train / 102 validation / 108 test). Config `v2` adds 650 training pairs and an 88-row `operator_test`.
	- [gesh75/netops-intent2cli-qwen0.5b-lora](https://huggingface.co/gesh75/netops-intent2cli-qwen0.5b-lora): experimental Qwen2.5-0.5B LoRA trained on v2. Adapter-only exact match: 12/108 (11.1%) on the frozen v1 test, 9/88 (10.2%) on `operator_test`. Optional retrieval+validator lookup mode (same card): 63/108 and 58/88 when the gold command is already in this corpus; it cannot return a command that is missing from the base. Not production-safe. Browser demo: [Intent finder](https://gesh75-multivendor-cli-configurator.static.hf.space/intent-finder.html) (retrieval only, no model).

	## Load

	```python
	from datasets import load_dataset
	ds = load_dataset("gesh75/multivendor-cli-commands", split="train")
	bgp = ds.filter(lambda r: r["vendor"] == "Juniper" and r["category"] == "BGP")
	```

	```python
	import pandas as pd
	df = pd.read_parquet("hf://datasets/gesh75/multivendor-cli-commands/data/train.parquet")
	```

	---

	### GESH · NetOps tools on Hugging Face

	\| \| \|
	\|---\|---\|
	\| 📡 [netlog-ai](https://huggingface.co/spaces/gesh75/netlog-ai) \| Sanitize-first network log analysis — browser lite demo of [netlog-ai](https://pypi.org/project/netlog-ai/) (PyPI 2.1.0) \|
	\| 🛠️ [multivendor-cli-configurator](https://huggingface.co/spaces/gesh75/multivendor-cli-configurator) \| 70k+ CLI commands across 17 vendors — cheatsheet, N-vendor compare, CLI Studio, Intent finder \|
	\| 🗂️ [multivendor-cli-commands](https://huggingface.co/datasets/gesh75/multivendor-cli-commands) \| The same 70,006 commands as a Parquet dataset \|
	\| 🧭 [netops-intent2cli](https://huggingface.co/datasets/gesh75/netops-intent2cli) \| v1 + v2 synthetic operator intents → corpus command pairs \|
	\| 🤖 [netops-intent2cli-qwen0.5b-lora](https://huggingface.co/gesh75/netops-intent2cli-qwen0.5b-lora) \| LoRA adapter: English intent → vendor CLI (experimental) \|

	Collection: [NetOps tools](https://huggingface.co/collections/gesh75/netops-tools-6aef9999054238474f968bcd) · Code: [github.com/gesh75](https://github.com/gesh75) · Docs hub: [gesh75.github.io](https://gesh75.github.io/)

	Doctrine: fail closed · sanitize first · evidence leaves, packets don't · honest coverage.