87 lines
2.5 KiB
Groff
87 lines
2.5 KiB
Groff
.TH HF_INFO 1 "November 2026" "nvfp4_gguf_quant" "User Commands"
|
|
.SH NAME
|
|
hf_info \- print architecture, quantization, and tensor layout from Hugging Face checkpoints
|
|
.SH SYNOPSIS
|
|
.B hf_info
|
|
[\fB\-\-metadata\fR]
|
|
[\fB\-\-tensors\fR]
|
|
[\fB\-h\fR|\fB\-\-help\fR]
|
|
\fImodel\fR [\fImodel\fR ...]
|
|
.SH DESCRIPTION
|
|
.B hf_info
|
|
inspects one or more Hugging Face model directories or standalone \fI.safetensors\fR files.
|
|
It reads configuration files (\fIconfig.json\fR, \fIgeneration_config.json\fR,
|
|
\fItokenizer_config.json\fR) and safetensors file headers without loading weight bytes
|
|
into RAM, allowing rapid inspection of multi-gigabyte or sharded checkpoints.
|
|
.PP
|
|
By default,
|
|
.B hf_info
|
|
displays a comprehensive summary, including:
|
|
.IP \(bu 2
|
|
Model architecture class, text/vision config hierarchy, and model type.
|
|
.IP \(bu 2
|
|
Model dimensions: context length, hidden size, intermediate size, layer counts, attention heads, and key-value heads.
|
|
.IP \(bu 2
|
|
Parameter counts: total parameters, active parameters, and size on disk.
|
|
.IP \(bu 2
|
|
Chat template configuration and capability detection (such as tool/function calling support).
|
|
.IP \(bu 2
|
|
Quantization format detection: Hugging Face quantization configs (bitsandbytes, AWQ, GPTQ, FP8),
|
|
embedded NVFP4 scales, and ComfyUI NVFP4 metadata (\fI_quantization_metadata\fR, \fIweight_scale\fR, \fI.comfy_quant\fR).
|
|
.IP \(bu 2
|
|
Sharding structure: list of shard files, header sizes, and index verification against \fImodel.safetensors.index.json\fR.
|
|
.SH OPTIONS
|
|
.TP
|
|
.BR \-h ", " \-\-help
|
|
Show command-line usage help and exit.
|
|
.TP
|
|
.B \-\-metadata
|
|
Print complete config, generation configuration, tokenizer configuration, and safetensors header metadata.
|
|
Long strings and large arrays are shortened for readability.
|
|
.TP
|
|
.B \-\-tensors
|
|
Print every tensor name, data type (e.g. \fIBF16\fR, \fIF16\fR, \fIU8\fR), shape, byte size,
|
|
and the shard file in which it resides.
|
|
.SH EXAMPLES
|
|
.TP
|
|
.B Inspect a Hugging Face model directory:
|
|
.RS
|
|
.nf
|
|
hf_info ./gemma-4-12B-it
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B Inspect a standalone safetensors file:
|
|
.RS
|
|
.nf
|
|
hf_info ./model-00001-of-00004.safetensors
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B Dump full config and chat template metadata:
|
|
.RS
|
|
.nf
|
|
hf_info \-\-metadata ./meta-llama-3-8b-instruct
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B List all tensors across all checkpoint shards:
|
|
.RS
|
|
.nf
|
|
hf_info \-\-tensors ./mistral-7b-v0.3
|
|
.fi
|
|
.RE
|
|
.SH EXIT STATUS
|
|
.TP
|
|
.B 0
|
|
Success.
|
|
.TP
|
|
.B 1
|
|
An error occurred (e.g. model directory not found or missing configuration).
|
|
.SH SEE ALSO
|
|
.BR gguf_info (1),
|
|
.BR hf_to_nvfp4_gguf (1),
|
|
.BR hf_to_nvfp4_safetensors (1),
|
|
.BR safetensors_to_nvfp4_gguf (1)
|
|
|