## Quant mix recipe created using Thireus' GGUF Tool Suite - https://gguf.thireus.com/

## Model head & embeddings — qbits: 32 6 5 
^output_norm\.weight$=f32
^token_embd\.weight$=iq5_ks
^output\.weight$=iq6_k

## Multi-headed attention parameters — qbits: 32 8 5 
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_norm\.weight$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_k_norm\.weight$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_v\.weight$=q8_0
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_v\.bias$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_q\.bias$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_q_norm\.weight$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_k\.bias$=f32
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_k\.weight$=q8_0
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_q\.weight$=iq5_ks
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.attn_output\.weight$=iq5_ks

## Dense Feed-Forward Network weights — qbits: 8 
^blk\.[0-2]\.ffn_gate\.weight$=q8_0
^blk\.[0-2]\.ffn_down\.weight$=q8_0
^blk\.[0-2]\.ffn_up\.weight$=q8_0

## NextN tensors — qbits: 32 5 
^blk\.92\.nextn\.enorm\.weight$=f32
^blk\.92\.nextn\.eh_proj\.weight$=iq5_ks
^blk\.92\.nextn\.shared_head_norm\.weight$=f32
^blk\.92\.nextn\.hnorm\.weight$=f32

## MoE Gating & Routing — qbits: 32 
^blk\.([3-9]|[1-8][0-9]|9[0-2])\.ffn_gate_inp\.weight$=f32
^blk\.([3-9]|[1-8][0-9]|9[0-2])\.exp_probs_b\.bias$=f32

## Misc / Other tensors — qbits: 32 
^blk\.([0-9]|[1-8][0-9]|9[0-2])\.post_attention_norm\.weight$=f32

## GPU-loaded - MoE Shared Experts Feed-Forward Network - ffn_*_shexp
# ffn_down_shexp — down-projection (shared experts) — qbits: 8 
^blk\.([3-9]|[1-8][0-9]|9[0-2])\.ffn_down_shexp\.weight$=q8_0

# ffn_up_shexp — up-projection (shared experts) — qbits: 8 
^blk\.([3-9]|[1-8][0-9]|9[0-2])\.ffn_up_shexp\.weight$=q8_0

# ffn_gate_shexp — gating network (shared experts) — qbits: 8 
^blk\.([3-9]|[1-8][0-9]|9[0-2])\.ffn_gate_shexp\.weight$=q8_0

## CPU-friendly - MoE Per-expert Feed-Forward Network - ffn_*_exps
# ffn_down_exps — down-projection (per-expert) — qbits: 4 3 2 
^blk\.(4[4-9]|5[0-9]|92)\.ffn_down_exps\.weight$=iq4_kss
^blk\.(18|2[4-5]|27|29|30|32|37|39|4[0-3]|6[0-9]|7[0-5])\.ffn_down_exps\.weight$=iq3_ks
^blk\.([3-9]|1[0-7]|19|2[0-3]|26|28|31|3[3-6]|38|7[6-9]|8[0-9]|9[0-1])\.ffn_down_exps\.weight$=iq2_kl

# ffn_up_exps — up-projection (per-expert) — qbits: 4 3 2 
^blk\.(47|49|5[2-6]|5[8-9]|6[2-5]|6[7-9]|70|73|92)\.ffn_up_exps\.weight$=iq4_kss
^blk\.(15|24|2[7-8]|30|34|3[6-9]|4[0-6]|48|[5-6][0-1]|57|66|7[1-2]|7[4-9]|9[0-1])\.ffn_up_exps\.weight$=iq3_ks
^blk\.([3-9]|1[0-4]|1[6-9]|2[0-3]|2[5-6]|29|3[1-3]|35|8[0-9])\.ffn_up_exps\.weight$=iq2_kl

# ffn_gate_exps — gating network (per-expert) — qbits: 4 3 2 
^blk\.(55|65|6[8-9]|9[1-2])\.ffn_gate_exps\.weight$=iq4_kss
^blk\.(25|2[7-8]|30|4[2-9]|[5-6][0-4]|5[6-9]|6[6-7]|7[0-9]|90)\.ffn_gate_exps\.weight$=iq3_ks
^blk\.([3-9]|1[0-9]|2[0-4]|26|29|3[1-9]|4[0-1]|8[0-9])\.ffn_gate_exps\.weight$=iq2_kl

## Summary of tensor sizes per class
# GPU Total: 12.13 GiB (100.0%) | 12.13 GiB max, if all were q8_0 | 12.13 GiB min, if all were q8_0
# CPU Total: 121.87 GiB (77.0%) | 158.24 GiB max, if all were iq4_kss | 106.90 GiB min, if all were iq2_kl
# GPU+CPU Total: 134.01 GiB (88.5%)

## Summary of tensor counts and bpw per qtype
#
# GPU-loaded quants:
# QTYPE		Count	BPW	Assigned GiB	% Assigned	Max GiB (all)
# +f32       	835	32    	  0.28 GiB	-		-
# q8_0      	465	8.5   	  3.63 GiB	100.0%		3.63
# +iq6_k     	1  	6.625 	  0.60 GiB	-		-
# +iq5_ks    	187	5.25  	  7.63 GiB	-		-
#
# CPU-friendly quants:
# QTYPE		Count	BPW	Assigned GiB	% Assigned	Max GiB (all)
# +iq5_ks    	1  	5.25  	  0.03 GiB	-		-
# +iq4_kss   	3  	4     	  1.76 GiB	-		-
# iq4_kss   	39 	4     	 22.85 GiB	14.6%		156.45
# iq3_ks    	102	3.1875	 47.63 GiB	38.2%		124.67
# iq2_kl    	126	2.6875	 49.60 GiB	47.2%		105.11
#
# -Average BPW: 3.2263
#
# -Notes:
# - '+' means user-defined pre-assigned tensors, or tensor missing from csv data or f32 tensors
# - Recipe produced on the 2025-11-27 04:19:33 WIB+0700 using Thireus' GGUF tools (https://gguf.thireus.com/)
# - Script SHA-256: 569b7f6a3239c9173d71ca1fadf34222607d72a2cfed2c284b42633e95b4a627
# - Calibration dataset 'models/GLM-4.6/kld_results.csv' SHA-256: c9eba4144dcd4c837fa68f3ad0620d43b9caafeec5928cfb3198034622b9166e
# - tensors.bf16.map SHA-256: fa987db60ed8e9eb4348a45cb0ad630f81a97b20292ed9adfe0369ddd3ec2828
# - tensors.bf16.map model name: GLM-4.6-THIREUS-BF16-SPECIAL_TENSOR-01760-of-01760
# - tensors.q8_0.map SHA-256: 4cdc6924a3e79e3a8df15cc40d607d01ad2d29375eb411ca423a44d437ae0c66
# - tensors.q8_0.map model name: GLM-4.6-THIREUS-Q8_0-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq6_k.map SHA-256: 0743f08065ebeeac64c52844c0a1cbde4e4e9242230d4218de7647fd46d2ba99
# - tensors.iq6_k.map model name: GLM-4.6-THIREUS-IQ6_K-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq5_ks.map SHA-256: 71bc70300215e17e1956fffd56bb0b63baaa6c09839333fa1ad59d6a47dd9455
# - tensors.iq5_ks.map model name: GLM-4.6-THIREUS-IQ5_KS-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq4_kss.map SHA-256: 78d8ec2bd1bd29d56add18b1685add51d29d4f953b2b6fb27f541e33009c73f4
# - tensors.iq4_kss.map model name: GLM-4.6-THIREUS-IQ4_KSS-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq3_ks.map SHA-256: 19804948e1f1bf32def27f7f4879a25a1bb22b2061720831c0841f940217bd01
# - tensors.iq3_ks.map model name: GLM-4.6-THIREUS-IQ3_KS-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq2_kl.map SHA-256: a13547ff8e7dcbc5ea25bdd506a8b224c8eef625d8a67f45b27b9e28ed3f7ea0
# - tensors.iq2_kl.map model name: GLM-4.6-THIREUS-IQ2_KL-SPECIAL_TENSOR-01760-of-01760
# - tensors.iq1_kt.map SHA-256: 793fce90c8b4c735e406f9ee701fc10d9e483d90fbdafdab533096ac7d9e748e
# - tensors.iq1_kt.map model name: GLM-4.6-THIREUS-IQ1_KT-SPECIAL_TENSOR-01760-of-01760
# - GPG signatures: DISABLED
# - Command used:
# ./quant_assign.py models/GLM-4.6/kld_results.csv --tolerance 0.001 --cpu-tensors-max-size 121.8 \
# --gpu-tensors-max-size 12.2 --exponential-factor 1.0 --skip-gpg --cpu-tensors \
# 'blk\.([3-9]|[1-8][0-9]|9[012])\.ffn_down_exps\.weight' 'blk\.([3-9]|[1-8][0-9]|9[012])\.ffn_up_exps\.weight' \
# 'blk\.([3-9]|[1-8][0-9]|9[012])\.ffn_gate_exps\.weight' --gpu-tensors '.*' --cpu-quants iq4_kss iq3_ks iq2_kl \
# --gpu-quants q8_0 --cpu-assign-tensors '^blk\.(92)\.ffn_down_exps\.weight=iq4_kss' \
# '^blk\.(92)\.ffn_up_exps\.weight=iq4_kss' '^blk\.(92)\.ffn_gate_exps\.weight=iq4_kss' \
# '^blk\.92\.nextn\.shared_head_head\.weight=iq5_ks' '^blk\.92\.nextn\.embed_tokens\.weight=iq5_ks' \
# '^blk\.92\.nextn\.eh_proj\.weight=iq5_ks' --gpu-assign-qtype iq5_ks --gpu-assign-tensors \
# '^blk\..*\.attn_(q|output)\.weight=iq5_ks' '^token_embd\.weight=iq5_ks' '^output\.weight=iq6_k' --harmonize-tensors \
# '' --harmonization-technique 0

## THE END!