No Python, no problem! Check the names of the layers of a safetensors easily.

So, the stupid question "does my LoRA have Text Encoder layers?" is often answered by using Python and like 3 lines of code. But not everyone has Python nor the will to install it.

That's ok, here is a quick and dirty binary to check that 😅

https://github.com/nArn0/read-st

The binaries are in the release, i compiled it for Windows, Linux and both Intel and Metal MacOSX. All around 2MB 👌

You can check the code if you want to see the uglyness of this quickly made bit of Golang:

package main

import (
	"encoding/binary"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"os"
	"sort"

	"github.com/nlpodyssey/safetensors"
	"golang.org/x/term"
)

func check(err error) {
	if err != nil {
		panic(err)
	}
}

func waitForEnter() {
   // Ugly mess of code to display a prompt and wait for Enter keypress
   // without displaying user input if other keys are pressed
	oldState, err := term.MakeRaw(int(os.Stdin.Fd()))
	check(err)
	defer term.Restore(int(os.Stdin.Fd()), oldState)
	screen := struct {
		io.Reader
		io.Writer
	}{os.Stdin, os.Stderr}
	terminal := term.NewTerminal(screen, "")
	_, _ = terminal.ReadPassword("Press 'Enter' to exit...")
}

func main() {
	// Check if argument is provided
	if len(os.Args) == 1 {
		panic(errors.New("no argument provided"))
	}

	// Open file provided as argument
	f, err := os.Open(os.Args[1])
	check(err)
	defer f.Close()

	// Read header size
	var l uint64
	err = binary.Read(f, binary.LittleEndian, &l)
	check(err)

	// Read header
	header := make([]byte, l)
	_, err = f.ReadAt(header, 8)
	check(err)

	// Get metadata from header
	var metadata safetensors.Metadata
	err = json.Unmarshal(header, &metadata)
	check(err)

	// Get tensors infos
	tensors := metadata.Tensors()

	// Get all tensors names
	layers := []string{}
	for name, _ := range tensors {
		layers = append(layers, name)
	}

	// Sort tensors names
	sort.Strings(layers)

	// Print names
	for _, name := range layers {
		fmt.Println(name)
	}

	// Wait for Enter keypress
	waitForEnter()
}

For usage, it's simple:

read-st-windows-amd64.exe nArnima.safetensors
lora_unet_blocks_0_adaln_modulation_cross_attn_1.alpha
lora_unet_blocks_0_adaln_modulation_cross_attn_1.lora_down.weight
lora_unet_blocks_0_adaln_modulation_cross_attn_1.lora_up.weight
lora_unet_blocks_0_adaln_modulation_cross_attn_2.alpha
lora_unet_blocks_0_adaln_modulation_cross_attn_2.lora_down.weight
lora_unet_blocks_0_adaln_modulation_cross_attn_2.lora_up.weight
(...)
lora_unet_blocks_9_self_attn_k_proj.alpha
lora_unet_blocks_9_self_attn_k_proj.lora_down.weight
lora_unet_blocks_9_self_attn_k_proj.lora_up.weight
lora_unet_blocks_9_self_attn_output_proj.alpha
lora_unet_blocks_9_self_attn_output_proj.lora_down.weight
lora_unet_blocks_9_self_attn_output_proj.lora_up.weight
lora_unet_blocks_9_self_attn_q_proj.alpha
lora_unet_blocks_9_self_attn_q_proj.lora_down.weight
lora_unet_blocks_9_self_attn_q_proj.lora_up.weight
lora_unet_blocks_9_self_attn_v_proj.alpha
lora_unet_blocks_9_self_attn_v_proj.lora_down.weight
lora_unet_blocks_9_self_attn_v_proj.lora_up.weight
Press 'Enter' to exit...

Even opening a terminal is not needed. On Windows, just drag and drop the safetensors file on top of the binary and you'll get what you want 😁

PS: i'll probably do a V2 later to get also the metadata and if this is a LoRA, the dimension. And also, with a bit more of check, for now, better not try it with something else than a safetensors file 😝