uploading IQ2_KL

README.md

@@ -31,7 +31,7 @@ Finally, I *really* appreciate the support from [aifoundry.org](https://aifoundr
 ## Quant Collection
 Perplexity computed against *wiki.test.raw*. (lower is "better")
 
-![Perplexity Chart](images/perplexity.png "Chart showing Perplexity vs Model Size.") TODO
+![Perplexity Chart](images/perplexity.png "Chart showing Perplexity vs Model Size.")
 
 These two are just test quants for baseline perplexity comparison and not available for download here:
 * `BF16` 1404.406 GiB (16.003 BPW)
@@ -111,6 +111,76 @@ numactl -N ${SOCKET} -m ${SOCKET} \
 
 </details>
 
+## IQ2_KL 261.988 GiB (2.985 BPW)
+PPL over 565 chunks for n_ctx=512 = 3.0217 +/- 0.01651
+
+NOTE: Actual used RAM/VRAM will be about 255.84 GiB despite larger model size reported due to unused blk.78/indexer/nextn tensors.
+
+<details>
+
+<summary>👈 Secret Recipe</summary>
+
+```bash
+#!/usr/bin/env bash
+
+custom="
+# 79 Repeating Layers [0-78]
+
+## Attention [0-78]
+blk\..*\.attn_k_b\.weight=q8_0
+blk\..*\.attn_v_b\.weight=q8_0
+blk\..*\.attn_kv_a_mqa\.weight=q8_0
+blk\..*\.attn_q_a\.weight=iq6_k
+blk\..*\.attn_q_b\.weight=iq6_k
+blk\..*\.attn_output\.weight=iq6_k
+
+# First 3 Dense Layers [0-2]
+blk\..*\.ffn_down\.weight=iq5_ks
+blk\..*\.ffn_(gate|up)\.weight=iq5_ks
+
+# Shared Expert Layers [3-78]
+blk\..*\.ffn_down_shexp\.weight=iq5_ks
+blk\..*\.ffn_(gate|up)_shexp\.weight=iq5_ks
+
+# Routed Experts Layers [3-78]
+# NOTE: blk.78.* NOT implemented at time of quantizing so no imatrix data available
+blk\.(78)\.ffn_down_exps\.weight=iq5_ks
+blk\.(78)\.ffn_(gate|up)_exps\.weight=iq5_ks
+blk\..*\.ffn_down_exps\.weight=iq3_ks
+blk\..*\.ffn_(gate|up)_exps\.weight=iq2_kl
+
+# Lightning indexer tensors [0-78]
+# NOTE: indexer.* NOT implemented at time of quantizing so no imatrix data available
+blk\..*\.indexer\.proj\.weight=q8_0
+blk\..*\.indexer\.attn_k\.weight=q8_0
+blk\..*\.indexer\.attn_q_b\.weight=iq6_k
+
+# NextN MTP Layer [78]
+# NOTE: nextn.* NOT implemented at time of quantizing so no imatrix data available
+blk\..*\.nextn\.eh_proj\.weight=q8_0
+
+# 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 ${SOCKET} -m ${SOCKET} \
+./build/bin/llama-quantize \
+    --custom-q "$custom" \
+    --imatrix /mnt/data/models/ubergarm/GLM-5-GGUF/imatrix-GLM-5-BF16.dat \
+    /mnt/data/models/ubergarm/GLM-5-GGUF/GLM-256x22B-5-BF16-00001-of-00033.gguf \
+    /mnt/data/models/ubergarm/GLM-5-GGUF/GLM-5-IQ2_KL.gguf \
+    IQ2_KL \
+    128
+```
+
+</details>
+
 ## smol-IQ2_KS 205.738 GiB (2.344 BPW)
 PPL over 565 chunks for n_ctx=512 = 3.7792 +/- 0.02183