Add IQ2_KL

README.md

@@ -217,6 +217,49 @@ numactl -N 0 -m 0 \
 
 </details>
 
+## `IQ2_KL` 169.597 GiB (3.034 BPW)
+Final estimate: PPL = 5.4113 +/- 0.03516
+
+<details>
+
+<summary>👈 Secret Recipe</summary>
+
+Originally this had issues with NaNs but got it working this this PR: https://github.com/ikawrakow/ik_llama.cpp/pull/735
+
+```bash
+#!/usr/bin/env bash
+
+custom="
+# Attention
+blk\..*\.attn_q.*=iq6_k
+blk\..*\.attn_k.*=q8_0
+blk\..*\.attn_v.*=q8_0
+blk\..*\.attn_output.*=iq6_k
+
+# Routed Experts
+blk\..*\.ffn_down_exps\.weight=iq3_k
+blk\..*\.ffn_(gate|up)_exps\.weight=iq2_kl
+
+# Non-Repeating Layers
+token_embd\.weight=iq4_k
+output\.weight=iq6_k
+"
+
+custom=$(
+  echo "$custom" | grep -v '^#' | \
+  sed -Ez 's:\n+:,:g;s:,$::;s:^,::'
+)
+
+numactl -N 0 -m 0 \
+./build/bin/llama-quantize \
+    --custom-q "$custom" \
+    --imatrix /mnt/raid/models/ubergarm/Qwen3-Coder-480B-A35B-Instruct-GGUF/imatrix-Qwen3-Coder-480B-A35B-Instruct-Q8_0.dat \
+    /mnt/raid/models/ubergarm/Qwen3-Coder-480B-A35B-Instruct-GGUF/Qwen3-Coder-480B-A35B-Instruct-BF16-00001-of-00021.gguf \
+    /mnt/raid/models/ubergarm/Qwen3-Coder-480B-A35B-Instruct-GGUF/Qwen3-Coder-480B-A35B-Instruct-IQ2_KL.gguf \
+    IQ2_KL \
+    192
+```
+
 ## `IQ2_K` 144.640 GiB (2.588 BPW)
 Final estimate: PPL = 5.6578 +/- 0.03697