96 lines
2.7 KiB
Groff
96 lines
2.7 KiB
Groff
.TH GGUF_INFO 1 "November 2026" "nvfp4_gguf_quant" "User Commands"
|
|
.SH NAME
|
|
gguf_info \- print architecture, quantization, and tensor metadata from a GGUF header
|
|
.SH SYNOPSIS
|
|
.B gguf_info
|
|
[\fB\-\-metadata\fR]
|
|
[\fB\-\-tensors\fR]
|
|
[\fB\-h\fR|\fB\-\-help\fR]
|
|
\fIgguf\fR [\fIgguf\fR ...]
|
|
.SH DESCRIPTION
|
|
.B gguf_info
|
|
inspects one or more GGUF files (versions 2 and 3) by reading only their header
|
|
and tensor directories. The raw weight bytes are never loaded into memory,
|
|
making the inspection fast even for multi-gigabyte models.
|
|
.PP
|
|
By default,
|
|
.B gguf_info
|
|
displays a concise summary of the model, including:
|
|
.IP \(bu 2
|
|
GGUF version, file size, and total tensor and key-value counts.
|
|
.IP \(bu 2
|
|
Model architecture identifier (e.g., \fIllama\fR, \fIqwen2\fR, \fIgemma2\fR).
|
|
.IP \(bu 2
|
|
General metadata: model name, author, license, context length, embedding length,
|
|
attention head counts, layer counts, feed-forward dimension, and vocabulary size.
|
|
.IP \(bu 2
|
|
Quantization summary: file type classification (e.g., \fIMOSTLY_NVFP4\fR,
|
|
\fIMOSTLY_Q4_K_M\fR, \fIMOSTLY_BF16\fR) and an exact count and breakdown of tensor
|
|
quantization types found across the model.
|
|
.SH OPTIONS
|
|
.TP
|
|
.BR \-h ", " \-\-help
|
|
Show command-line usage help and exit.
|
|
.TP
|
|
.B \-\-metadata
|
|
Print every key-value metadata pair found in the GGUF header. Long string
|
|
values, large numeric arrays, and vocabulary/merges arrays are truncated or
|
|
summarized for readability.
|
|
.TP
|
|
.B \-\-tensors
|
|
Print every tensor present in the file with its full name, GGML quantization
|
|
type (e.g., \fINVFP4\fR, \fIBF16\fR, \fIF16\fR, \fIQ4_K\fR), shape dimensions,
|
|
and byte offset within the file.
|
|
.SH SUPPORTED QUANTIZATION TYPES
|
|
.B gguf_info
|
|
recognizes all standard GGML types including:
|
|
.PP
|
|
.BR F32 ", " F16 ", " BF16 ", " F64 ", "
|
|
.BR Q4_0 ", " Q4_1 ", " Q5_0 ", " Q5_1 ", " Q8_0 ", " Q8_1 ", "
|
|
.BR Q2_K ", " Q3_K ", " Q4_K ", " Q5_K ", " Q6_K ", " Q8_K ", "
|
|
.BR IQ1_S ", " IQ1_M ", " IQ2_XXS ", " IQ2_XS ", " IQ2_S ", "
|
|
.BR IQ3_XXS ", " IQ3_S ", " IQ4_XS ", " IQ4_NL ", "
|
|
.BR TQ1_0 ", " TQ2_0 ", " MXFP4 ", " NVFP4 ", " Q1_0 ", " Q2_0 .
|
|
.SH EXAMPLES
|
|
.TP
|
|
.B Inspect a single GGUF model:
|
|
.RS
|
|
.nf
|
|
gguf_info model-NVFP4.gguf
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B Dump all metadata keys and parameters:
|
|
.RS
|
|
.nf
|
|
gguf_info \-\-metadata model-NVFP4.gguf
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B List all tensors and their formats:
|
|
.RS
|
|
.nf
|
|
gguf_info \-\-tensors model-NVFP4.gguf
|
|
.fi
|
|
.RE
|
|
.TP
|
|
.B Inspect multiple models in batch:
|
|
.RS
|
|
.nf
|
|
gguf_info model-f16.gguf model-NVFP4.gguf
|
|
.fi
|
|
.RE
|
|
.SH EXIT STATUS
|
|
.TP
|
|
.B 0
|
|
Success.
|
|
.TP
|
|
.B 1
|
|
An error occurred (e.g. file not found, invalid magic bytes, or unsupported GGUF version).
|
|
.SH SEE ALSO
|
|
.BR hf_info (1),
|
|
.BR gguf_to_nvfp4_gguf (1),
|
|
.BR hf_to_nvfp4_gguf (1),
|
|
.BR ollama_nvfp4 (1)
|
|
|