Upload folder using huggingface_hub

.gitattributes

@@ -64,3 +64,4 @@ tinyllama-chat/gguf/model-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 gemma4-random-model/gguf-q4_0/gemma4-random-model-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3-random-model/gguf-q4_0/qwen3-random-model-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
 smollm3-random-model/gguf-q4_0/smollm3-random-model-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
+gemma4-random-model/gguf-q4_k/gemma4-random-model-Q4_K.gguf filter=lfs diff=lfs merge=lfs -text

ARCHITECTURE_RANDOM_MODELS_REPORT.json

@@ -6,14 +6,15 @@
       "eog_token_ids": [
         1
       ],
-      "gguf_quantization": "Q4_0",
-      "gguf_sha256": "2b631587994acd127d3e6fcc67552c707d6fdba22beada963597387b9e814881",
+      "gguf_quantization": "Q4_K",
+      "gguf_sha256": "3ce6309aa4221289441b198ab63cd6c0d03075ba183ea1990001313c99d6e845",
       "hf_dtype": "bfloat16",
-      "hf_sha256": "18c8b09759c831fbc9d6a8caf7df8e9d395508d4aec09336167f675527b3a360",
-      "parameter_count": 1519168,
+      "hf_sha256": "e97de7f7fcad67f15547487a4b44fa8b6066ad8cd49b2c0d4b1230e3cf615ad3",
+      "parameter_count": 6036608,
       "tensor_type_histogram": {
         "F32": 44,
-        "Q4_0": 42
+        "Q4_K": 37,
+        "Q6_K": 5
       },
       "validation": "passed"
     },

README.md

@@ -37,7 +37,7 @@ checkpoint weights are included.
 | `deepseek-coder` | Llama | 9,296 | F32 | Q4_K_M | linear RoPE scaling |
 | `hermes3-llama31` | Llama | 86,336 | F32 | Q4_K_M | Llama 3 RoPE scaling |
 | `livekit-turn-detector` | Llama | 132,336 | F32 | Q4_K_M | explicit head dimension, GQA |
-| `gemma4-random-model` | Gemma 4 | 1,519,168 | BF16 | Q4_0 | five-local/one-global attention schedule |
+| `gemma4-random-model` | Gemma 4 | 6,036,608 | BF16 | Q4_K | five-local/one-global attention schedule |
 | `qwen3-random-model` | Qwen 3 | 508,800 | BF16 | Q4_0 | wide Q projection and Q/K head norms |
 | `smollm3-random-model` | SmolLM3 | 4,917,504 | BF16 | Q4_0 | three-RoPE/one-no-RoPE layer schedule |
 
@@ -76,8 +76,8 @@ The architecture-specific cases use:
 |   |-- model.safetensors
 |   |-- tokenizer.json
 |   `-- tokenizer_config.json
-|-- gguf-q4_0/
-|   |-- <case>-Q4_0.gguf
+|-- gguf-q4_k/ or gguf-q4_0/
+|   |-- <case>-Q4_K.gguf or <case>-Q4_0.gguf
 |   `-- quantize.log
 |-- reference/
 |   |-- inputs.json
@@ -113,7 +113,7 @@ model EOS semantics. Per-case metadata records the actual tensor-type histogram,
 hashes, commands, and informational comparison with the corresponding
 Transformers reference.
 
-Q4_K_M and Q4_0 are lossy formats. Their logits are not required to equal the
+Q4_K, Q4_K_M, and Q4_0 are lossy formats. Their logits are not required to equal the
 F32 or BF16 reference exactly.
 
 ## Reproducibility and scope
@@ -130,3 +130,21 @@ repositories; consult their recorded provenance before redistribution.
 
 See `REPORT.md`, `GGUF_Q4_K_M_REPORT.json`, and
 `ARCHITECTURE_RANDOM_MODELS_REPORT.json` for collection-level summaries.
+
+## History
+
+### 2026-08-12: Gemma 4 random model rebuilt
+
+The first `gemma4-random-model` release used hidden width 128, 1,519,168
+parameters, and Q4_0. That version was replaced because its small matrix axes
+did not exercise K-quant blocks and its GGUF metadata was not sufficiently close
+to the inspected 12B Gemma 4 source GGUF.
+
+The current release uses hidden width 256, FFN width 1024, 6,036,608 parameters,
+and llama.cpp's `Q4_K` alias. Its actual tensor histogram contains F32, Q4_K,
+and Q6_K, matching the source profile family. It preserves the six-layer
+five-sliding/one-full schedule, per-layer KV head array, local/global head-width
+ratio, dual RoPE regimes, global shared K/V behavior, complete norm inventory,
+layer output scales, tied embeddings, tokenizer special IDs, and applicable
+sampling metadata. HF BF16 weights and both HF and GGUF references were
+regenerated; the previous Q4_0 Gemma 4 files are not part of this release.

SHA256SUMS

@@ -1,7 +1,7 @@
-c9732b21b282ff81879aa6eb2a61784a14825331847b62384da6d0a81f7218ea  ARCHITECTURE_RANDOM_MODELS_REPORT.json
+d17b37c112047f3ac47f54634c1a59da55e179b4fb176ad596ecc94240750006  ARCHITECTURE_RANDOM_MODELS_REPORT.json
 18ec0cdad98853d7f10381790713f0b3190bd4f24962594bf576e11fdb07f76f  GGUF_Q4_K_M_ADDED_FOUR_REPORT.json
 695036ee0e223d3d79170fb3a0f795eb88f6337fb2e29c18f0b549979c057208  GGUF_Q4_K_M_REPORT.json
-9c7a6588b7355830d75c57171a9368a7db87d192b0f05443db1233199cd893b6  README.md
+fb3008e24a0fecc2bb0cb72b27770db78f72bc481dc086b8080c89f09e419816  README.md
 95d67a407dc96839250fe23e85e0a5bab9d355ee463b0c87bf9e33164c952b52  REPORT.md
 02130c5ef4fe4e00bf26cc6e0eb288b90916acf827031bf572ffc02a48d5f3f0  deepseek-coder/case.json
 616662945b2382e3c4ef36c05b40ebacea17e2e16fa3a723b7fd49eb5d0e9fdd  deepseek-coder/config-diff.json
@@ -22,17 +22,18 @@ b80f4edf3204563fd78e3c95498aa4ceb9af1770f52869c638d33be9124a43a5  deepseek-coder
 603a68ffd4df8d0b01492da00fcd7b7386f7a8bbe90e92b81b82516a9231fbfb  deepseek-coder/tokenizer/tokenizer_config.json
 7b1258e1304c5f09f3bc64cd4b569e7ddfa55a062c0581d0e5cb1a3d7a17fcc8  deepseek-coder/tokenizer/vocabulary.json
 aa20ff35795ac323bbdbc81c292f0fa74bb81f9026b88cd01c40c4d81e428d8c  deepseek-coder/validation.json
-399fff213eafc2d462587c9031564b5aafcb99436fc965d9f4e6c8677b6f1837  gemma4-random-model/CONFIG_DECISION.md
-2b631587994acd127d3e6fcc67552c707d6fdba22beada963597387b9e814881  gemma4-random-model/gguf-q4_0/gemma4-random-model-Q4_0.gguf
-5b733665980cfba8e31508035a6088b0e53f9a09c68bd0fcb0f53721bb140f1e  gemma4-random-model/gguf-q4_0/quantize.log
-7386dc7fbb3fc1e83a7db585d088c646183c8c700581bfc12b95390a515c6ba1  gemma4-random-model/hf-bf16/config.json
-f0ce61128449f564ff1dccf316e54c54429f5c5ae9a398803de39d452a671e85  gemma4-random-model/hf-bf16/generation_config.json
-18c8b09759c831fbc9d6a8caf7df8e9d395508d4aec09336167f675527b3a360  gemma4-random-model/hf-bf16/model.safetensors
-8538827f687d1fd46c3de2ba983d0af5e705c0b0661bab1e8f261a9372046d4d  gemma4-random-model/hf-bf16/tokenizer.json
-729e9569f612365498a64b12bf1c2fc8816b8b56cba7be3c5841ab1b0a346a7f  gemma4-random-model/hf-bf16/tokenizer_config.json
-d5f3c821f738e7cea4c8a9abb9d8283f168b23e67cec7528dfaa84f7be48094f  gemma4-random-model/metadata.json
-ceb65d119a54448e1d7ebbb5431284979f33ce309966b3149f667dc4caa24911  gemma4-random-model/reference/gguf-native.json
-ffa41b6c27bab211af7994632041ddedf47c460ad6b89482909d793604831d38  gemma4-random-model/reference/hf-outputs.safetensors
+4d463e831b4df46925f7ed82f898a9212d004882dacefc7012171254a70e52ed  gemma4-random-model/CONFIG_DECISION.md
+5c330d499055016556a88ee42b398fac5fa6fa7706f68cad2c4aa8e6d7286f1c  gemma4-random-model/gguf-q4_k/convert.log
+3ce6309aa4221289441b198ab63cd6c0d03075ba183ea1990001313c99d6e845  gemma4-random-model/gguf-q4_k/gemma4-random-model-Q4_K.gguf
+762c5f1a0d751251c0a73f75b629f7256f94dda7431f3a5520b36068e24f4ea1  gemma4-random-model/gguf-q4_k/quantize.log
+571a499253eb22afcc050ee3428fdc56dcc23a38cfc66248a911bc2b51df93bb  gemma4-random-model/hf-bf16/config.json
+47a105f62d5f925c70d1b8d2825d617c034bdbab7ed99aadaaf35ac56751e13f  gemma4-random-model/hf-bf16/generation_config.json
+e97de7f7fcad67f15547487a4b44fa8b6066ad8cd49b2c0d4b1230e3cf615ad3  gemma4-random-model/hf-bf16/model.safetensors
+5497f30f2a945c00a2d2cbe8b2f37a8a7c27a596fa01f59c61eef3a55bdf4a4c  gemma4-random-model/hf-bf16/tokenizer.json
+d0fab7832fecaac8554eaebfad305d55655cc79b3404eb9a7b0a4fc9dce08a01  gemma4-random-model/hf-bf16/tokenizer_config.json
+76212c94e1734b8c71f4cbd4949d351cfcbc7dce2392a8b3331ae9c17829385f  gemma4-random-model/metadata.json
+3f3f233fc1b56d7a63862da867fa81e37c78fc4f1a169386dd01a5cfa09fc28a  gemma4-random-model/reference/gguf-native.json
+d6bf946d748543d85bb840ccf2533c4ed30817adec2e736e2f27e7024e3f7996  gemma4-random-model/reference/hf-outputs.safetensors
 719a2f5ea966e92e42ebcc43dd73f0714b5d740c9f426d22c942dbb1e7efb6ab  gemma4-random-model/reference/inputs.json
 94a8d82acf2fde73ae5095df8f17c0e2303df4510d0836f3c5c6d65ed51444e5  hermes3-llama31/case.json
 6c0670415f473f79fed118e88160012de23a21df169ed664c2f56be817aa755d  hermes3-llama31/config-diff.json
@@ -72,7 +73,7 @@ f003d32e36ddc9d7c7a50c186da5c544b2828a0d04ca62aeac741d765727a635  livekit-turn-d
 564a5f0d3380d83d0ffcfa509a5e5369c0610f586cb6b3d0d45ccaee76610327  livekit-turn-detector/tokenizer/tokenizer_config.json
 7b1258e1304c5f09f3bc64cd4b569e7ddfa55a062c0581d0e5cb1a3d7a17fcc8  livekit-turn-detector/tokenizer/vocabulary.json
 391312ee129392e2b35b6a238593ca5ddff3dd51c6c324135f1bb0114be1e068  livekit-turn-detector/validation.json
-853c059a6216e5edc98ad1f444fb4dd010d263ebef36a7e5ca3a3ff715c3ab36  manifest.json
+d3c192b83c78f0d46d46835f0e4e201dd80755dde5fd39016a051b5133b36410  manifest.json
 047c4545232dc055c2c7e4a30a70c22e0e75e9e777e539732128f65c32b0c1cb  minicpm5/case.json
 282a55aae55b59573109fe37222197f298600ad691573bcd84ea5d4dc1afcc0a  minicpm5/config-diff.json
 d088266a01cd2f9a6107877932d6af9f65e6d00202292226eab76af8ea0e889f  minicpm5/gguf/metadata.json

gemma4-random-model/CONFIG_DECISION.md

@@ -2,13 +2,13 @@
 
 ## Status
 
-The text-only tiny configuration is fixed and generated under
+The text-only reduced configuration is generated under
 `artifacts/gemma4-v0/gemma4-random-model` as a BF16 Hugging Face package and a
-Q4_0 GGUF fixture.
+mixed K-quant GGUF requested through llama.cpp's `Q4_K` alias.
 
-The fixed Hugging Face implementation is Transformers 5.14.1
+The current Hugging Face implementation is Transformers 5.15.0
 `Gemma4ForCausalLM` with `Gemma4TextConfig`. Its constructed trainable parameter
-count is 1,519,168.
+count is 6,036,608.
 
 The local source file identifies itself as `general.architecture = gemma4`.
 Repository and revision provenance are not inferable from the local directory and
@@ -39,13 +39,13 @@ and `general.type = mmproj`. It is not included in the first text-only fixture.
 
 ## Tiny geometry
 
-The selected geometry is defined in `configs/gemma4-tiny-v0.json`:
+The selected geometry is defined in `configs/gemma4-random-q4k-v1.json`:
 
 - 6 layers: one complete five-local/one-global schedule.
-- Hidden width 128 and FFN width 512.
+- Hidden width 256 and FFN width 1024.
 - 4 query heads.
-- Local layers: 2 KV heads, head width 32.
-- Global layer: 1 KV head, head width 64, no independent V projection.
+- Local layers: 2 KV heads, head width 64.
+- Global layer: 1 KV head, head width 128, no independent V projection.
 - Context 128 and sliding window 64.
 - Vocabulary 128 with PAD/EOS/BOS/UNK/MASK IDs 0/1/2/3/4.
 
@@ -54,9 +54,8 @@ preserves the official layer schedule, the 2:1 global/local head-width ratio, th
 2:1 query/local-KV head ratio, the single global KV head, FFN ratio 4, separate
 RoPE regimes, softcap, tied embeddings, and global shared-KV tensor inventory.
 
-All matrix dimensions are multiples of 32. This is required so that a direct F32
-GGUF can subsequently be quantized through the pinned Q4_0 path without changing
-model geometry merely to satisfy quantization blocks.
+All matrix quantization axes are multiples of 256. This allows actual Q4_K/Q6_K
+tensors rather than Q4_0 or K-quant fallback caused by undersized dimensions.
 
 ## Expected GGUF tensor dimensions
 
@@ -64,14 +63,14 @@ GGUF dimensions are listed in reader order.
 
 | Role | Local layers 0-4 | Global layer 5 |
 |---|---:|---:|
-| `attn_q.weight` | `[128, 128]` | `[128, 256]` |
-| `attn_k.weight` | `[128, 64]` | `[128, 64]` |
-| `attn_v.weight` | `[128, 64]` | absent |
-| `attn_output.weight` | `[128, 128]` | `[256, 128]` |
-| `attn_q_norm.weight` | `[32]` | `[64]` |
-| `attn_k_norm.weight` | `[32]` | `[64]` |
-
-Every layer also has FFN gate/up `[128, 512]`, FFN down `[512, 128]`,
+| `attn_q.weight` | `[256, 256]` | `[256, 512]` |
+| `attn_k.weight` | `[256, 128]` | `[256, 128]` |
+| `attn_v.weight` | `[256, 128]` | absent |
+| `attn_output.weight` | `[256, 256]` | `[512, 256]` |
+| `attn_q_norm.weight` | `[64]` | `[128]` |
+| `attn_k_norm.weight` | `[64]` | `[128]` |
+
+Every layer also has FFN gate/up `[256, 1024]`, FFN down `[1024, 256]`,
 hidden-width norms, and one scalar layer-output scale.
 
 ## Generation gates
@@ -83,5 +82,5 @@ Fixture generation is accepted only when:
 - a second generation is byte-identical;
 - the F32 GGUF loads in the pinned llama.cpp revision;
 - direct-token prefill and cached decode are finite and reproducible;
-- the Q4_0 conversion retains the same geometry and loads successfully;
+- the mixed Q4_K/Q6_K conversion retains the same geometry and loads successfully;
 - F32 outputs are compared with an independently generated matched reference.

gemma4-random-model/gguf-q4_k/convert.log

@@ -0,0 +1,137 @@
+INFO:hf-to-gguf:Loading model: hf-for-converter
+WARNING:hf-to-gguf:Failed to load model config from /home/codex/tmp/gemma4-q4k-work-20260812/hf-for-converter: The checkpoint you are trying to load has model type `gemma4_text_converter_raw_config` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.
+
+You can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`
+WARNING:hf-to-gguf:Trying to load config.json instead
+INFO:hf-to-gguf:Model architecture: Gemma4ForCausalLM
+WARNING:hf-to-gguf:Failed to load model config from /home/codex/tmp/gemma4-q4k-work-20260812/hf-for-converter: The checkpoint you are trying to load has model type `gemma4_text_converter_raw_config` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.
+
+You can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`
+WARNING:hf-to-gguf:Trying to load config.json instead
+INFO:hf-to-gguf:gguf: indexing model part 'model.safetensors'
+INFO:gguf.gguf_writer:gguf: This GGUF file is for Little Endian only
+INFO:hf-to-gguf:Exporting model...
+INFO:hf-to-gguf:rope_freqs.weight,                torch.float32 --> F32, shape = {64}
+INFO:hf-to-gguf:token_embd.weight,                torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.0.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.0.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.0.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.0.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.0.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.0.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.0.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.0.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.0.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.0.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.0.attn_output.weight,         torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.0.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.0.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.0.attn_v.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.1.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.1.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.1.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.1.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.1.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.1.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.1.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.1.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.1.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.1.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.1.attn_output.weight,         torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.1.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.1.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.1.attn_v.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.2.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.2.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.2.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.2.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.2.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.2.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.2.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.2.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.2.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.2.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.2.attn_output.weight,         torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.2.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.2.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.2.attn_v.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.3.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.3.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.3.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.3.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.3.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.3.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.3.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.3.attn_output.weight,         torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.3.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.3.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.3.attn_v.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.4.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.4.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.4.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.4.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.4.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.4.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.4.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.4.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.4.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.4.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.4.attn_output.weight,         torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.4.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {64}
+INFO:hf-to-gguf:blk.4.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 256}
+INFO:hf-to-gguf:blk.4.attn_v.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.5.attn_norm.weight,           torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.5.layer_output_scale.weight,  torch.bfloat16 --> F32, shape = {1}
+INFO:hf-to-gguf:blk.5.ffn_down.weight,            torch.bfloat16 --> F32, shape = {1024, 256}
+INFO:hf-to-gguf:blk.5.ffn_gate.weight,            torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.5.ffn_up.weight,              torch.bfloat16 --> F32, shape = {256, 1024}
+INFO:hf-to-gguf:blk.5.post_attention_norm.weight, torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.5.post_ffw_norm.weight,       torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.5.ffn_norm.weight,            torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:blk.5.attn_k_norm.weight,         torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.5.attn_k.weight,              torch.bfloat16 --> F32, shape = {256, 128}
+INFO:hf-to-gguf:blk.5.attn_output.weight,         torch.bfloat16 --> F32, shape = {512, 256}
+INFO:hf-to-gguf:blk.5.attn_q_norm.weight,         torch.bfloat16 --> F32, shape = {128}
+INFO:hf-to-gguf:blk.5.attn_q.weight,              torch.bfloat16 --> F32, shape = {256, 512}
+INFO:hf-to-gguf:output_norm.weight,               torch.bfloat16 --> F32, shape = {256}
+INFO:hf-to-gguf:Set meta model
+INFO:hf-to-gguf:Set model parameters
+INFO:hf-to-gguf:gguf: context length = 128
+INFO:hf-to-gguf:gguf: embedding length = 256
+INFO:hf-to-gguf:gguf: feed forward length = 1024
+INFO:hf-to-gguf:gguf: head count = 4
+INFO:hf-to-gguf:gguf: key-value head count = 2
+WARNING:hf-to-gguf:Unknown RoPE type: proportional
+INFO:hf-to-gguf:gguf: rope scaling type = NONE
+INFO:hf-to-gguf:gguf: rope theta = 1000000.0
+INFO:hf-to-gguf:gguf: rope theta swa = 10000.0
+INFO:hf-to-gguf:gguf: rms norm epsilon = 1e-06
+INFO:hf-to-gguf:gguf: file type = 0
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.context_length', overwriting it with new value 128 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.head_count', overwriting it with new value 4 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.layer_norm_rms_epsilon', overwriting it with new value 1e-06 of type FLOAT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.key_length', overwriting it with new value 64 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.value_length', overwriting it with new value 64 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.rope.freq_base', overwriting it with new value 1000000.0 of type FLOAT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.head_count_kv', overwriting it with new value 2 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.key_length', overwriting it with new value 128 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.value_length', overwriting it with new value 128 of type UINT32
+WARNING:gguf.gguf_writer:Duplicated key name 'gemma4.attention.head_count_kv', overwriting it with new value [2, 2, 2, 2, 2, 1] of type ARRAY
+INFO:hf-to-gguf:Set model quantization version
+INFO:hf-to-gguf:Set model tokenizer
+[transformers] You are using a model of type `gemma4_text_converter_raw_config` to instantiate a model of type ``. This may be expected if you are loading a checkpoint that shares a subset of the architecture (e.g., loading a `sam2_video` checkpoint into `Sam2Model`), but is otherwise not supported and can yield errors. Please verify that the checkpoint is compatible with the model you are instantiating.
+INFO:gguf.vocab:Adding 1 merge(s).
+INFO:gguf.vocab:Setting special token type bos to 2
+INFO:gguf.vocab:Setting special token type eos to 1
+INFO:gguf.vocab:Setting special token type unk to 3
+INFO:gguf.vocab:Setting special token type pad to 0
+INFO:gguf.vocab:Setting special token type mask to 4
+INFO:gguf.gguf_writer:Writing the following files:
+INFO:gguf.gguf_writer:/home/codex/tmp/gemma4-q4k-work-20260812/gemma4-random-model-F32.gguf: n_tensors = 86, total_size = 24.1M
+
+Writing:   0%|          | 0.00/24.1M [00:00<?, ?byte/s]
+Writing: 100%|██████████| 24.1M/24.1M [00:00<00:00, 983Mbyte/s]
+INFO:hf-to-gguf:Model successfully exported to /home/codex/tmp/gemma4-q4k-work-20260812/gemma4-random-model-F32.gguf

gemma4-random-model/gguf-q4_k/gemma4-random-model-Q4_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ce6309aa4221289441b198ab63cd6c0d03075ba183ea1990001313c99d6e845
+size 3591488

gemma4-random-model/gguf-q4_k/quantize.log

@@ -0,0 +1,138 @@
+llama_print_build_info: build = 0 (unknown)
+llama_print_build_info: built with Clang 21.1.8 for Linux x86_64
+llama_quantize: quantizing '/home/codex/tmp/gemma4-q4k-work-20260812/gemma4-random-model-F32.gguf' to '/home/codex/conf_track/artifacts/gemma4-v0/gemma4-random-model/gguf-q4_k/gemma4-random-model-Q4_K.gguf' as Q4_K
+llama_model_loader: loaded meta data with 41 key-value pairs and 86 tensors from /home/codex/tmp/gemma4-q4k-work-20260812/gemma4-random-model-F32.gguf (version GGUF V3 (latest))
+llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
+llama_model_loader: - kv   0:                       general.architecture str              = gemma4
+llama_model_loader: - kv   1:                               general.type str              = model
+llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 64
+llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
+llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
+llama_model_loader: - kv   5:                               general.name str              = Gemma 4 Random Model
+llama_model_loader: - kv   6:                         general.size_label str              = 6.0M
+llama_model_loader: - kv   7:                         gemma4.block_count u32              = 6
+llama_model_loader: - kv   8:                      gemma4.context_length u32              = 128
+llama_model_loader: - kv   9:                    gemma4.embedding_length u32              = 256
+llama_model_loader: - kv  10:                 gemma4.feed_forward_length u32              = 1024
+llama_model_loader: - kv  11:                gemma4.attention.head_count u32              = 4
+llama_model_loader: - kv  12:             gemma4.attention.head_count_kv arr[i32,6]       = [2, 2, 2, 2, 2, 1]
+llama_model_loader: - kv  13:                      gemma4.rope.freq_base f32              = 1000000.000000
+llama_model_loader: - kv  14:                  gemma4.rope.freq_base_swa f32              = 10000.000000
+llama_model_loader: - kv  15:    gemma4.attention.layer_norm_rms_epsilon f32              = 0.000001
+llama_model_loader: - kv  16:                gemma4.attention.key_length u32              = 128
+llama_model_loader: - kv  17:              gemma4.attention.value_length u32              = 128
+llama_model_loader: - kv  18:                          general.file_type u32              = 0
+llama_model_loader: - kv  19:             gemma4.final_logit_softcapping f32              = 30.000000
+llama_model_loader: - kv  20:            gemma4.attention.sliding_window u32              = 64
+llama_model_loader: - kv  21:          gemma4.attention.shared_kv_layers u32              = 0
+llama_model_loader: - kv  22:    gemma4.embedding_length_per_layer_input u32              = 0
+llama_model_loader: - kv  23:    gemma4.attention.sliding_window_pattern arr[bool,6]      = [true, true, true, true, true, false]
+llama_model_loader: - kv  24:            gemma4.attention.key_length_swa u32              = 64
+llama_model_loader: - kv  25:          gemma4.attention.value_length_swa u32              = 64
+llama_model_loader: - kv  26:                gemma4.rope.dimension_count u32              = 128
+llama_model_loader: - kv  27:            gemma4.rope.dimension_count_swa u32              = 64
+llama_model_loader: - kv  28:               general.quantization_version u32              = 2
+llama_model_loader: - kv  29:                       tokenizer.ggml.model str              = gemma4
+llama_model_loader: - kv  30:                      tokenizer.ggml.tokens arr[str,128]     = ["<pad>", "<eos>", "<bos>", "<unk>", ...
+llama_model_loader: - kv  31:                      tokenizer.ggml.scores arr[f32,128]     = [-1000.000000, -1000.000000, -1000.00...
+llama_model_loader: - kv  32:                  tokenizer.ggml.token_type arr[i32,128]     = [3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, ...
+llama_model_loader: - kv  33:                      tokenizer.ggml.merges arr[str,1]       = ["a b"]
+llama_model_loader: - kv  34:                tokenizer.ggml.bos_token_id u32              = 2
+llama_model_loader: - kv  35:                tokenizer.ggml.eos_token_id u32              = 1
+llama_model_loader: - kv  36:            tokenizer.ggml.unknown_token_id u32              = 3
+llama_model_loader: - kv  37:            tokenizer.ggml.padding_token_id u32              = 0
+llama_model_loader: - kv  38:               tokenizer.ggml.mask_token_id u32              = 4
+llama_model_loader: - kv  39:            tokenizer.ggml.add_space_prefix bool             = false
+llama_model_loader: - kv  40:               tokenizer.ggml.add_bos_token bool             = true
+llama_model_loader: - type  f32:   86 tensors
+[   1/  86] output_norm.weight                   - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[   2/  86] rope_freqs.weight                    - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[   3/  86] token_embd.weight                    - [   256,    128,      1,      1], type =    f32, converting to q6_K .. size =     0.12 MiB ->     0.03 MiB
+[   4/  86] blk.0.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[   5/  86] blk.0.attn_k_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[   6/  86] blk.0.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[   7/  86] blk.0.attn_output.weight             - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[   8/  86] blk.0.attn_q.weight                  - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[   9/  86] blk.0.attn_q_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  10/  86] blk.0.attn_v.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  11/  86] blk.0.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  12/  86] blk.0.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  13/  86] blk.0.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  14/  86] blk.0.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  15/  86] blk.0.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  16/  86] blk.0.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  17/  86] blk.0.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  18/  86] blk.1.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  19/  86] blk.1.attn_k_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  20/  86] blk.1.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  21/  86] blk.1.attn_output.weight             - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  22/  86] blk.1.attn_q.weight                  - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  23/  86] blk.1.attn_q_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  24/  86] blk.1.attn_v.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  25/  86] blk.1.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  26/  86] blk.1.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  27/  86] blk.1.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  28/  86] blk.1.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  29/  86] blk.1.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  30/  86] blk.1.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  31/  86] blk.1.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  32/  86] blk.2.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  33/  86] blk.2.attn_k_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  34/  86] blk.2.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  35/  86] blk.2.attn_output.weight             - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  36/  86] blk.2.attn_q.weight                  - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  37/  86] blk.2.attn_q_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  38/  86] blk.2.attn_v.weight                  - [   256,    128,      1,      1], type =    f32, converting to q6_K .. size =     0.12 MiB ->     0.03 MiB
+[  39/  86] blk.2.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q6_K .. size =     1.00 MiB ->     0.21 MiB
+[  40/  86] blk.2.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  41/  86] blk.2.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  42/  86] blk.2.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  43/  86] blk.2.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  44/  86] blk.2.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  45/  86] blk.2.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  46/  86] blk.3.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  47/  86] blk.3.attn_k_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  48/  86] blk.3.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  49/  86] blk.3.attn_output.weight             - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  50/  86] blk.3.attn_q.weight                  - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  51/  86] blk.3.attn_q_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  52/  86] blk.3.attn_v.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  53/  86] blk.3.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  54/  86] blk.3.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  55/  86] blk.3.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  56/  86] blk.3.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  57/  86] blk.3.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  58/  86] blk.3.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  59/  86] blk.3.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  60/  86] blk.4.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  61/  86] blk.4.attn_k_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  62/  86] blk.4.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  63/  86] blk.4.attn_output.weight             - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  64/  86] blk.4.attn_q.weight                  - [   256,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.25 MiB ->     0.04 MiB
+[  65/  86] blk.4.attn_q_norm.weight             - [    64,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  66/  86] blk.4.attn_v.weight                  - [   256,    128,      1,      1], type =    f32, converting to q6_K .. size =     0.12 MiB ->     0.03 MiB
+[  67/  86] blk.4.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  68/  86] blk.4.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  69/  86] blk.4.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  70/  86] blk.4.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  71/  86] blk.4.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  72/  86] blk.4.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  73/  86] blk.4.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  74/  86] blk.5.attn_k.weight                  - [   256,    128,      1,      1], type =    f32, converting to q4_K .. size =     0.12 MiB ->     0.02 MiB
+[  75/  86] blk.5.attn_k_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  76/  86] blk.5.attn_norm.weight               - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  77/  86] blk.5.attn_output.weight             - [   512,    256,      1,      1], type =    f32, converting to q4_K .. size =     0.50 MiB ->     0.07 MiB
+[  78/  86] blk.5.attn_q.weight                  - [   256,    512,      1,      1], type =    f32, converting to q4_K .. size =     0.50 MiB ->     0.07 MiB
+[  79/  86] blk.5.attn_q_norm.weight             - [   128,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  80/  86] blk.5.ffn_down.weight                - [  1024,    256,      1,      1], type =    f32, converting to q6_K .. size =     1.00 MiB ->     0.21 MiB
+[  81/  86] blk.5.ffn_gate.weight                - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  82/  86] blk.5.ffn_norm.weight                - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  83/  86] blk.5.ffn_up.weight                  - [   256,   1024,      1,      1], type =    f32, converting to q4_K .. size =     1.00 MiB ->     0.14 MiB
+[  84/  86] blk.5.layer_output_scale.weight      - [     1,      1,      1,      1], type =    f32, size =    0.000 MiB
+[  85/  86] blk.5.post_attention_norm.weight     - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+[  86/  86] blk.5.post_ffw_norm.weight           - [   256,      1,      1,      1], type =    f32, size =    0.001 MiB
+llama_model_quantize_impl: model size  =    23.03 MiB (32.00 BPW)
+llama_model_quantize_impl: quant size  =     3.42 MiB (4.75 BPW)
+
+llama_quantize: quantize time =    88.65 ms
+llama_quantize:    total time =    88.65 ms

gemma4-random-model/hf-bf16/config.json

@@ -10,13 +10,12 @@
   "enable_moe_block": false,
   "eos_token_id": 1,
   "final_logit_softcapping": 30.0,
-  "global_head_dim": 64,
-  "head_dim": 32,
+  "head_dim": 64,
   "hidden_activation": "gelu_pytorch_tanh",
-  "hidden_size": 128,
+  "hidden_size": 256,
   "hidden_size_per_layer_input": 0,
   "initializer_range": 0.02,
-  "intermediate_size": 512,
+  "intermediate_size": 1024,
   "layer_types": [
     "sliding_attention",
     "sliding_attention",
@@ -30,11 +29,16 @@
   "moe_intermediate_size": null,
   "num_attention_heads": 4,
   "num_experts": null,
-  "num_global_key_value_heads": 1,
   "num_hidden_layers": 6,
   "num_key_value_heads": 2,
   "num_kv_shared_layers": 0,
   "pad_token_id": 0,
+  "per_layer_config": {
+    "5": {
+      "head_dim": 128,
+      "num_key_value_heads": 1
+    }
+  },
   "rms_norm_eps": 1e-06,
   "rope_parameters": {
     "full_attention": {
@@ -50,10 +54,10 @@
   "sliding_window": 64,
   "tie_word_embeddings": true,
   "top_k_experts": null,
-  "transformers_version": "5.14.1",
+  "transformers_version": "5.15.0",
   "use_bidirectional_attention": null,
   "use_cache": true,
   "use_double_wide_mlp": false,
   "vocab_size": 128,
-  "vocab_size_per_layer_input": 128
+  "vocab_size_per_layer_input": 0
 }

gemma4-random-model/hf-bf16/generation_config.json

@@ -2,9 +2,9 @@
   "_from_model_config": true,
   "bos_token_id": 2,
   "eos_token_id": 1,
-  "output_attentions": false,
-  "output_hidden_states": false,
   "pad_token_id": 0,
-  "transformers_version": "5.14.1",
-  "use_cache": true
+  "temperature": 1.0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.15.0"
 }

gemma4-random-model/hf-bf16/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18c8b09759c831fbc9d6a8caf7df8e9d395508d4aec09336167f675527b3a360
-size 3047372
+oid sha256:e97de7f7fcad67f15547487a4b44fa8b6066ad8cd49b2c0d4b1230e3cf615ad3
+size 12082356

gemma4-random-model/hf-bf16/tokenizer.json

@@ -51,7 +51,10 @@
   ],
   "normalizer": null,
   "pre_tokenizer": {
-    "type": "Whitespace"
+    "type": "ByteLevel",
+    "add_prefix_space": false,
+    "trim_offsets": true,
+    "use_regex": true
   },
   "post_processor": {
     "type": "TemplateProcessing",
@@ -79,9 +82,29 @@
     ],
     "special_tokens": {}
   },
-  "decoder": null,
+  "decoder": {
+    "type": "Sequence",
+    "decoders": [
+      {
+        "type": "ByteFallback"
+      },
+      {
+        "type": "ByteLevel",
+        "add_prefix_space": true,
+        "trim_offsets": true,
+        "use_regex": true
+      }
+    ]
+  },
   "model": {
-    "type": "WordLevel",
+    "type": "BPE",
+    "dropout": null,
+    "unk_token": "<unk>",
+    "continuing_subword_prefix": null,
+    "end_of_word_suffix": null,
+    "fuse_unk": false,
+    "byte_fallback": true,
+    "ignore_merges": false,
     "vocab": {
       "<pad>": 0,
       "<eos>": 1,
@@ -212,6 +235,11 @@
       "<t126>": 126,
       "<t127>": 127
     },
-    "unk_token": "<unk>"
+    "merges": [
+      [
+        "a",
+        "b"
+      ]
+    ]
   }
 }

gemma4-random-model/hf-bf16/tokenizer_config.json

@@ -3,7 +3,7 @@
   "bos_token": "<bos>",
   "eos_token": "<eos>",
   "mask_token": "<mask>",
-  "model_max_length": 128,
+  "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<pad>",
   "tokenizer_class": "TokenizersBackend",
   "unk_token": "<unk>"

gemma4-random-model/metadata.json

@@ -2,70 +2,85 @@
   "architecture": "gemma4",
   "gguf": {
     "comparison_to_hf_bf16": {
-      "decode": {
-        "max_abs_error": 0.15816828608512878,
-        "mean_abs_error": 0.046988509595394135,
-        "top1_agreement": true,
-        "top5_overlap": 3
+      "decode_logits": {
+        "max_absolute_error": 0.317632794380188,
+        "mean_absolute_error": 0.09652481973171234,
+        "top1_agreement": false,
+        "top5_overlap": 1
       },
-      "policy": "informational because Q4_0 is lossy",
-      "prefill_last": {
-        "max_abs_error": 0.15831683576107025,
-        "mean_abs_error": 0.03859168291091919,
+      "policy": "informational; mixed K-quant and BF16 executions are not required to be identical",
+      "prefill_last_logits": {
+        "max_absolute_error": 0.25260016322135925,
+        "mean_absolute_error": 0.07460260391235352,
         "top1_agreement": true,
         "top5_overlap": 3
       }
     },
+    "converter_command": [
+      "PYTHON",
+      "CONVERTER",
+      "HF_FOR_CONVERTER",
+      "--outfile",
+      "F32_OUTPUT",
+      "--outtype",
+      "f32",
+      "--model-name",
+      "Gemma 4 Random Model"
+    ],
     "direct_token_validation": {
       "all_logits_finite": true,
       "decode_token_id": 17,
       "eog_token_ids": [
         1
       ],
-      "exactly_reproducible": true,
       "prefill_token_ids": [
         5,
         7,
         11,
         13
-      ]
+      ],
+      "repeated_output_byte_identical": true,
+      "status": "passed"
     },
-    "f32_intermediate_retained": false,
-    "f32_intermediate_sha256": "634ea1838571d13516f8615d5be35cd4d2b35845e3e70445244971c4fa63ff24",
-    "model_sha256": "2b631587994acd127d3e6fcc67552c707d6fdba22beada963597387b9e814881",
-    "model_size_bytes": 877088,
-    "quantization": "Q4_0",
+    "f32_intermediate_retained_in_package": false,
+    "f32_intermediate_sha256": "67b5f94511c78c951619397ea719f1fb0de81bafacc01b00690d507fb677b57a",
+    "llama_cpp_commit": "0b1bad14ff204627636aeb1de22ddcd5acb859d4",
+    "llama_cpp_profile": "Q4_K_M alias",
+    "model_sha256": "3ce6309aa4221289441b198ab63cd6c0d03075ba183ea1990001313c99d6e845",
+    "model_size_bytes": 3591488,
+    "pure": false,
     "quantizer_command": [
       "LLAMA_QUANTIZE",
-      "--token-embedding-type",
-      "q4_0",
+      "OVERRIDE_SOURCE_LIKE_METADATA",
       "F32_INPUT",
       "OUTPUT",
-      "Q4_0"
+      "Q4_K"
     ],
+    "requested_quantization": "Q4_K",
     "tensor_type_histogram": {
       "F32": 44,
-      "Q4_0": 42
+      "Q4_K": 37,
+      "Q6_K": 5
     }
   },
   "hf": {
-    "config_sha256": "7386dc7fbb3fc1e83a7db585d088c646183c8c700581bfc12b95390a515c6ba1",
+    "config_sha256": "571a499253eb22afcc050ee3428fdc56dcc23a38cfc66248a911bc2b51df93bb",
     "construct_prefill_decode_reload": "passed",
     "dtype": "bfloat16",
-    "model_sha256": "18c8b09759c831fbc9d6a8caf7df8e9d395508d4aec09336167f675527b3a360",
+    "model_sha256": "e97de7f7fcad67f15547487a4b44fa8b6066ad8cd49b2c0d4b1230e3cf615ad3",
     "reload_tolerance": {
       "atol": 0.02,
       "rtol": 0.02
     },
-    "tokenizer_json_sha256": "8538827f687d1fd46c3de2ba983d0af5e705c0b0661bab1e8f261a9372046d4d",
-    "transformers_version": "5.14.1"
+    "tokenizer_json_sha256": "5497f30f2a945c00a2d2cbe8b2f37a8a7c27a596fa01f59c61eef3a55bdf4a4c",
+    "transformers_version": "5.15.0"
   },
-  "parameter_count": 1519168,
+  "parameter_count": 6036608,
   "reproducibility": {
-    "gguf_q4_0": true,
-    "hf_config": true,
-    "hf_model": true,
-    "second_generation_byte_identical": true
+    "gguf_q4_k_byte_identical": true,
+    "hf_config_byte_identical": true,
+    "hf_model_byte_identical": true,
+    "native_repeated_output_byte_identical": true
   },
   "rng": {
     "increment": 1442695040888963407,
@@ -73,6 +88,34 @@
     "name": "tlfloat LCG64 equation",
     "warmup_steps": 10
   },
-  "schema_version": 1,
-  "seed": 5135595944287666177
+  "schema_version": 2,
+  "seed": 5135595944287666177,
+  "source_gguf": {
+    "architecture": "gemma4",
+    "intentionally_changed_or_omitted": {
+      "geometry": "reduced while preserving ratios and one full schedule",
+      "identity": "random-model identity replaces upstream model identity",
+      "tokenizer.chat_template": "omitted because the 128-token synthetic vocabulary lacks its control tokens",
+      "tokenizer.ggml.suppress_tokens": "omitted because source IDs 258882 and 258883 are outside the reduced vocabulary"
+    },
+    "metadata_field_count": 45,
+    "path": "/home/codex/llm_models/gguf/gemma-4-12B-it-GGUF/gemma-4-12B-it-Q4_K_M.gguf",
+    "preserved_metadata": [
+      "sampling defaults",
+      "six-layer attention schedule",
+      "RoPE regimes",
+      "RMS epsilon",
+      "final logit softcap",
+      "tokenizer model and special IDs",
+      "add BOS",
+      "no added space prefix",
+      "quantization version"
+    ],
+    "tensor_count": 667,
+    "tensor_type_histogram": {
+      "F32": 338,
+      "Q4_K": 284,
+      "Q6_K": 45
+    }
+  }
 }

gemma4-random-model/reference/gguf-native.json

@@ -5,6 +5,6 @@
   "decode_token_id": 17,
   "eog_token_ids": [1],
   "all_logits_finite": true,
-  "prefill_last_logits": [-0.243439302,0.0599161685,-0.0845743716,0.0522050597,0.0797347426,-0.03387703,-0.0461352468,-0.238350868,0.0519622341,0.0766301826,0.126724079,-0.0139343906,-0.0291910302,0.118146136,0.120272741,-0.0258229449,-0.159660712,-0.00222994783,0.00440907292,-0.0304125454,0.0980525315,0.0801480561,-0.00244864542,0.142669708,0.015994221,-0.0381400324,-0.0229879357,-0.0514277704,-0.0138719492,0.273642391,0.218537062,0.126278937,0.070124574,-0.0690698773,0.031898085,-0.0267212726,-0.0291228201,-0.0225784294,-0.239331439,0.167331189,-0.127066836,0.221266046,-0.126179755,0.206429392,-0.143929034,-0.0102543291,0.229470193,-0.0172765926,0.0780417323,-0.0969515741,0.0769872814,0.0613183305,0.0403772257,-0.168282136,0.231201753,0.0205171276,0.0230379496,-0.209012553,-0.0319907963,-0.0245357733,0.00959140435,0.0801036358,-0.0883303657,0.0970800146,-0.115716472,-0.0724429861,-0.124726042,0.124294072,-0.0134264585,0.0358970277,0.227650881,0.00189068564,0.155431747,-0.244728908,0.0816485435,0.0228621587,0.139080092,0.173921034,-0.100639582,-0.0593391173,-0.0474654809,0.0911222324,0.121439792,-0.149972886,-0.208900452,-0.188849151,0.218660593,0.0517382175,-0.197950065,-0.169730186,-0.163051397,0.113981888,0.172181055,-0.00614864333,0.115805477,-0.262914896,-0.132733464,0.0576047525,-0.108596526,-0.162751198,-0.0619868003,0.0415232368,-0.157174632,-0.00733210798,-0.12342795,-0.137874156,-0.101731159,0.0428587683,-0.10437981,0.0370034501,-0.0460586175,0.0158869345,-0.0467884727,0.101396501,-0.235209301,-0.0206604954,-0.0808326676,0.109388076,0.191693529,-0.0848258659,-0.00442003552,-0.0363867059,-0.154983073,-0.014337346,0.274131835,-0.263285935,0.0940570235,0.0207084212],
-  "decode_logits": [-0.00950608216,-0.0892616585,-0.0102732191,-0.0360522121,0.110710457,0.0217158236,0.0268691275,0.11116679,0.062429022,0.14466773,0.0287577808,0.120581679,-0.0712457374,-0.112511128,0.105950512,-0.0633386895,-0.0917795077,-0.0265769325,0.0225345306,-0.25535512,0.167570621,0.112365395,0.0230123922,0.0484296866,-0.228892609,-0.147982344,-0.116228707,-0.0728008151,-0.128445566,-0.0128700966,-0.0407301299,0.113672934,0.128408462,-0.0344361551,0.020843599,-0.107788987,-0.0234028604,0.213245809,-0.0919425413,0.0751381218,-0.253096074,0.262579083,-0.150305837,0.243521363,-0.145598575,-0.0906658098,0.0682541281,0.0127086695,0.0152440453,-0.0776171535,-0.0459475666,-0.00190381811,-0.148405954,-0.173953682,0.126784325,-0.213859275,0.116676956,-0.0121320002,-0.127416849,-0.0586256646,0.0533445515,0.010526645,-0.42138359,0.0762371942,-0.0866874307,0.181853861,-0.14425239,-0.0253368318,-0.152625233,0.0170733649,0.0313455909,0.0770332366,0.0380382352,-0.154644251,0.0201064609,0.0427017771,0.01267628,0.108827092,-0.219167039,-0.0583654344,-0.0495231189,0.0334515274,0.0151894297,0.0707492009,-0.0993220806,-0.126987964,0.0974186882,0.0565137677,-0.355304003,-0.134302884,-0.0124985855,0.104582824,0.145622939,-0.123666488,-0.0822865516,-0.383201808,-0.0574632362,-0.0576772653,-0.117835112,0.0597717986,-0.0567533672,0.122403659,-0.207931876,-0.062526606,-0.132351458,0.157630011,0.0877739787,-0.0715343878,-0.0917522386,-0.135613829,0.0618881881,-0.139158815,-0.0728474706,0.0418399647,0.147925481,0.122347273,-0.109589078,-0.0333428718,0.0310138203,-0.167800277,-0.00797175709,-0.00486898189,-0.0671501011,0.136077702,-0.0265318714,-0.110943288,-0.0150498049,0.147386327]
+  "prefill_last_logits": [-0.180541143,-0.0625896752,-0.0801645443,-0.0324444957,-0.145822316,0.16689758,-0.056453649,0.0354969651,-0.107032984,-0.0642904267,0.278022289,0.138497159,-0.381877869,0.114222273,-0.18239069,-0.0383097045,0.0305028781,0.173735902,0.0596304536,0.18742837,0.148728207,-0.0641550347,-0.0438717008,-0.135932371,-0.0398499295,-0.12629649,-0.0909607634,0.290424764,0.00942425523,-0.124148756,-0.0854420438,-0.0250708275,0.0894175395,0.0316273794,-0.0333787538,-0.0799205527,-0.0867133439,-0.101149172,0.211644024,-0.191840932,-0.0421341397,0.285777003,-0.116382837,-0.247553647,-0.0722068548,-0.0523612201,-0.334899545,0.0216170233,0.438813061,0.289526731,-0.210583016,-0.0481212623,-0.172922224,0.337610394,0.0481857546,-0.0580814593,0.172419965,-0.306156754,0.222406477,-0.285389692,0.0797131285,0.166056708,0.188024804,-0.445678234,0.0883253366,-0.20347251,-0.268209398,0.373594314,0.232743144,-0.069143571,-0.0496469587,-0.16557017,-0.0130885988,-0.0726044923,-0.171762243,0.206093356,0.120467141,0.0200306084,0.21257633,-0.627123415,0.00390911149,0.0121024838,-0.421131998,-0.0978042036,-0.139752686,-0.140550151,-0.0453886986,-0.136265546,-0.0397286192,-0.121006839,-0.272427201,-0.213812858,-0.0442257412,-0.679793835,-0.475235343,-0.13458927,-0.0113198459,0.308448523,-0.0260743145,0.0897181258,0.358040959,-0.105817452,-0.205231041,-0.0239973925,-0.222660467,0.0930277854,-0.170702145,-0.00103891653,-0.0503437631,-0.0191276055,-0.275645614,-0.0427258126,-0.379553288,-0.398210794,0.0957396328,0.193684846,-0.107572451,-0.030029377,-0.0339817703,0.209344536,0.135555431,0.101219699,0.282283813,-0.0226774644,0.0111512542,-0.116329782,0.00453716516,0.189975291],
+  "decode_logits": [0.0986062139,0.103315219,-0.0652525946,0.123670734,-0.0419863798,-0.314382672,0.0190353524,0.288026989,0.179623559,0.00199305289,-0.131245092,-0.0432820022,-0.228420675,-0.205682784,-0.198196515,-0.153465882,-0.234333411,-0.0483023711,0.220053941,-0.0592789762,-0.132185638,-0.00524511933,0.168493405,-0.094419606,0.0581862815,-0.259100229,0.12117058,0.100616656,-0.0987538546,-0.209920093,0.0236835964,-0.00266891741,0.199776173,-0.0309187006,-0.260020614,0.404630393,-0.200569421,0.0688071623,0.0932103917,-0.323904306,0.248116791,-0.0643995851,-0.278670967,-0.37724492,-0.0890700072,-0.185746729,0.0648792908,0.105546869,0.0685760975,-0.163680673,0.022758726,-0.265505493,-0.0260222349,-0.0517332368,0.153072178,0.141724363,0.123938538,-0.198548824,0.0563444607,0.0641695559,0.288197339,-0.0911298022,0.232513264,-0.361397266,-0.153648838,-0.0139489193,-0.0716145262,0.239581585,-0.0615705475,0.0815294012,0.0684768707,0.0789057314,0.108183503,0.0321343988,-0.12308383,0.332421839,-0.312578231,-0.157823935,0.032510329,0.00258246087,0.19458437,0.133237913,0.0968403295,-0.141325116,-0.0631495714,0.116116427,-0.147599086,0.0742732137,0.163263127,-0.112877861,0.105354428,-0.312680155,-0.302541375,-0.140019819,-0.0941601098,0.0485069491,-0.0505474918,-0.122098878,0.303493977,-0.117373735,0.298803478,-0.295070529,0.264631271,0.124640383,-0.00693055242,0.0502441116,-0.27799198,0.292767376,0.0496037453,0.32225278,0.144747049,0.135104164,-0.14159067,-0.335766494,-0.158481583,0.158381671,0.0570854172,-0.00177340209,-0.105467319,0.239013657,0.276010066,0.0803631246,0.564333439,0.29532069,-0.066697076,-0.181127667,-0.0701091066,-0.33402738]
 }

gemma4-random-model/reference/hf-outputs.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ffa41b6c27bab211af7994632041ddedf47c460ad6b89482909d793604831d38
+oid sha256:d6bf946d748543d85bb840ccf2533c4ed30817adec2e736e2f27e7024e3f7996
 size 2720

manifest.json

@@ -137,19 +137,19 @@
       "architecture": "gemma4",
       "auxiliary_tokenizer": {
         "path": "gemma4-random-model/hf-bf16/tokenizer.json",
-        "sha256": "8538827f687d1fd46c3de2ba983d0af5e705c0b0661bab1e8f261a9372046d4d"
+        "sha256": "5497f30f2a945c00a2d2cbe8b2f37a8a7c27a596fa01f59c61eef3a55bdf4a4c"
       },
       "case": "gemma4-random-model",
       "packages": {
-        "gguf_q4_0": {
-          "path": "gemma4-random-model/gguf-q4_0/gemma4-random-model-Q4_0.gguf",
-          "sha256": "2b631587994acd127d3e6fcc67552c707d6fdba22beada963597387b9e814881",
-          "size_bytes": 877088
+        "gguf_q4_k": {
+          "path": "gemma4-random-model/gguf-q4_k/gemma4-random-model-Q4_K.gguf",
+          "sha256": "3ce6309aa4221289441b198ab63cd6c0d03075ba183ea1990001313c99d6e845",
+          "size_bytes": 3591488
         },
         "safetensors_bf16": {
           "path": "gemma4-random-model/hf-bf16/model.safetensors",
-          "sha256": "18c8b09759c831fbc9d6a8caf7df8e9d395508d4aec09336167f675527b3a360",
-          "size_bytes": 3047372
+          "sha256": "e97de7f7fcad67f15547487a4b44fa8b6066ad8cd49b2c0d4b1230e3cf615ad3",
+          "size_bytes": 12082356
         }
       }
     },