AesSedai/GLM-4.5-GGUF

This repository contains some custom quants of GLM-4.5 that focus on a couple of different schemas compared to the usual quantization schemas.

The idea being that given the huge size of the FFN tensors compared to the rest of the tensors in the model, it should be possible to achieve a better quality while keeping the overall size of the entire model smaller compared to a similar naive quantization.

The following charts showcase a very wide variety of GLM-4.5 quants that were tested for KL Divergence using the reference logits and corpus included in this repo: GLM-4.5-KLD-8192-ref-logits-ed-combined-all-micro-Q8_0.bin and combined_all_micro.txt

A full CSV with the data is included as well in glm-4.5-quantization-output.csv.

The naming convention is (inconsistently) as follows, generally: [Default Type]-[FFN_UP]-[FFN_GATE]-[FFN_DOWN], eg: Q6_K-IQ2_S-IQ2_S-IQ3_S. This means:

• Q6_K is the default type (attention, shared expert, etc.)
• IQ2_S was used for the FFN_UP and FFN_GATE conditional expert tensors
• IQ3_S was used for the FFN_DOWN conditional expert tensors

Generally speaking, quants following the above convention tend to have a better KLD and PPL compared to quants from other providers. Visualized here are the Mean KLD and Mean PPL of the quants on the Pareto frontier (so, best for a given size). Full graphs are available in the plots-glm-4.5-8192 folder.

Provided here are a few quants, separated into llama.cpp and ik_llama.cpp folders for convenience (though, ik_llama.cpp is capable of running the quants in llama.cpp, but the opposite is not true).

llama.cpp imatrix Quantizations of zai-org/GLM-4.5

This quant collection can be run on llama.cpp or kobold.cpp like normal.

Q6_K-IQ2_S-IQ2_S-IQ3_S: 128.18 GiB (3.07 BPW), Final estimate: PPL = 4.786993 ± 0.031213, KLD = 0.145117 ± 0.002232

GLM-4.5-Q6_K-Q2_K-Q2_K-Q3_K: 129.57 GiB (3.11 BPW), Final estimate: PPL = 4.700384 ± 0.030202, KLD = 0.164863 ± 0.002339

GLM-4.5-Q8_0-IQ3_XXS-IQ3_XXS-IQ3_S: 144.68 GiB (3.47 BPW), Final estimate: PPL = 4.729934 ± 0.030769, KLD = 0.116520 ± 0.002072

ik_llama.cpp imatrix Quantizations of zai-org/GLM-4.5

This quant collection REQUIRES ik_llama.cpp fork to support the ik's latest SOTA quants and optimizations! Do not download these big files and expect them to run on mainline vanilla llama.cpp, ollama, LM Studio, KoboldCpp, etc!

NOTE ik_llama.cpp can also run your existing GGUFs from bartowski, unsloth, mradermacher, etc if you want to try it out before downloading my quants.

Some of ik's new quants are supported with Nexesenex/croco.cpp fork of KoboldCPP with Windows builds for CUDA 12.9. Also check for Windows builds by Thireus here. which have been CUDA 12.8.

See Ubergarm's GLM-4.5 quants for info on how to use the recipe or make your own quant.

IQ2_KT: 109.269 GiB (2.619 BPW): Lost the results somewhere, oops.

👈 Recipe

# 93 Repeating Layers [0-92]

# Attention
blk\..*\.attn_q.*=iq4_k
blk\..*\.attn_k.*=iq6_k
blk\..*\.attn_v.*=iq6_k
blk\..*\.attn_output.*=iq5_ks

# First 3 Dense Layers [0-2]
blk\..*\.ffn_down\.weight=iq4_ks
blk\..*\.ffn_(gate|up)\.weight=iq3_ks

# Shared Expert Layers [3-92]
blk\..*\.ffn_down_shexp\.weight=iq6_k
blk\..*\.ffn_(gate|up)_shexp\.weight=iq6_k

# Routed Experts Layers [3-92]
blk\..*\.ffn_down_exps\.weight=iq3_kt
blk\..*\.ffn_(gate|up)_exps\.weight=iq2_kt

# NextN MTP Layer [92]
blk\..*\.nextn\.embed_tokens\.weight=iq4_k
blk\..*\.nextn\.shared_head_head\.weight=iq6_k
blk\..*\.nextn\.eh_proj\.weight=iq6_k

# Non-Repeating Layers
token_embd\.weight=iq4_k
output\.weight=iq6_k

IQ4_KSS: 176.499 GiB (4.231 BPW): Lost the results somewhere, oops.

👈 Recipe

# 93 Repeating Layers [0-92]

# Attention
blk\.(0|1|2)\.attn_q.*=q8_0
blk\.(0|1|2)\.attn_k.*=q8_0
blk\.(0|1|2)\.attn_v.*=q8_0
blk\.(0|1|2)\.attn_output.*=q8_0

blk\..*\.attn_q.*=iq6_k
blk\..*\.attn_k.*=iq6_k
blk\..*\.attn_v.*=iq6_k
blk\..*\.attn_output.*=iq6_k

# First 3 Dense Layers [0-2]
blk\..*\.ffn_down\.weight=iq5_ks
blk\..*\.ffn_(gate|up)\.weight=iq4_ks

# Shared Expert Layers [3-92]
blk\..*\.ffn_down_shexp\.weight=q8_0
blk\..*\.ffn_(gate|up)_shexp\.weight=q8_0

# Routed Experts Layers [3-92]
blk\..*\.ffn_down_exps\.weight=iq4_ks
blk\..*\.ffn_(gate|up)_exps\.weight=iq4_kss

# NextN MTP Layer [92]
blk\..*\.nextn\.embed_tokens\.weight=iq5_ks
blk\..*\.nextn\.shared_head_head\.weight=iq5_ks
blk\..*\.nextn\.eh_proj\.weight=q8_0

# Non-Repeating Layers
token_embd\.weight=iq4_k
output\.weight=iq6_k

IQ4_KS-IQ4_KS-IQ5_KS: 200.326 GiB (4.802 BPW), Final estimate: PPL = 4.618597 ± 0.029981, KLD = 0.072590 ± 0.001816

👈 Recipe

Default quant level @ Q8_0

# Shared Expert Layers [3-92]
blk\..*\.ffn_down_shexp\.weight=q8_0
blk\..*\.ffn_(gate|up)_shexp\.weight=q8_0

# Routed Experts Layers [3-92]
blk\..*\.ffn_up_exps\.weight=iq4_ks
blk\..*\.ffn_gate_exps\.weight=iq4_ks
blk\..*\.ffn_down_exps\.weight=iq5_ks

IQ5_K: 204.948 GiB (4.913 BPW), Final estimate: PPL = 4.665419 ± 0.030393, KLD = 0.078092 ± 0.001891

👈 Recipe

# 93 Repeating Layers [0-92]

# Attention
blk\.(0|1|2)\.attn_q.*=q8_0
blk\.(0|1|2)\.attn_k.*=q8_0
blk\.(0|1|2)\.attn_v.*=q8_0
blk\.(0|1|2)\.attn_output.*=q8_0

blk\..*\.attn_q.*=iq5_k
blk\..*\.attn_k.*=iq5_k
blk\..*\.attn_v.*=iq5_k
blk\..*\.attn_output.*=iq5_k

# First 3 Dense Layers [0-2]
blk\..*\.ffn_down\.weight=q8_0
blk\..*\.ffn_(gate|up)\.weight=q8_0

# Shared Expert Layers [3-92]
blk\..*\.ffn_down_shexp\.weight=q8_0
blk\..*\.ffn_(gate|up)_shexp\.weight=q8_0

# Routed Experts Layers [3-92]
blk\..*\.ffn_down_exps\.weight=iq5_k
blk\..*\.ffn_(gate|up)_exps\.weight=iq4_k

# NextN MTP Layer [92]
blk\..*\.nextn\.embed_tokens\.weight=iq5_k
blk\..*\.nextn\.shared_head_head\.weight=iq5_k
blk\..*\.nextn\.eh_proj\.weight=q8_0

# Non-Repeating Layers
token_embd\.weight=q8_0
output\.weight=q8_0