Upload README.md with huggingface_hub

README.md

@@ -1,533 +1,42 @@
 ---
-license: apache-2.0
-base_model:
-- google/gemma-4-31B-it
-- MRockatansky/Gemma-4-31B-storymaxxed
-- MRockatansky/Gemma-4-31B-storymaxxed2
-- LatitudeGames/Equinox-31B
-- sophosympatheia/Mero-Artemis-31B-v0.3.1
-- AuriAetherwiing/G4-31B-Musica-v1
-- llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-- Jackrong/Gemopus-4-31B-it
-- p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
+base_model: Blazed-Forge/Gemma-4-Giftige-Blume-31B-v1
+library_name: transformers
 tags:
-- merge
-- mergekit
-- reasoning
-- non-reasoning
-- creative writing
-- roleplay
-- uncensored
-- nsfw
-- 31B
-- gemma-4
+- quantized
+- w8a16
+- fp8
 ---
 
-<style>
-.blume-wrapper {
-background-color: #050806;
-background-image: radial-gradient(circle at 10% 10%, rgba(168, 85, 247, 0.08) 0%, transparent 40%), radial-gradient(circle at 90% 90%, rgba(34, 197, 94, 0.06) 0%, transparent 40%);
-color: #cbd5e1;
-font-family: 'Segoe UI', Roboto, Helvetica, Arial, sans-serif;
-padding: 40px;
-border-radius: 25px;
-border: 1px solid #14532d;
-line-height: 1.7;
-position: relative;
-overflow: hidden;
-box-shadow: inset 0 0 100px rgba(0,0,0,0.8), 0 20px 50px rgba(0,0,0,0.6);
-}
-.toxic-header { text-align: center; padding-bottom: 30px; position: relative; z-index: 10; }
-.toxic-title {
-font-size: 3.8rem;
-font-weight: 800;
-margin: 15px 0 5px 0;
-background: linear-gradient(90deg, #c522af, #c6f755, #c522af);
--webkit-background-clip: text;
--webkit-text-fill-color: transparent;
-filter: drop-shadow(0 0 15px rgba(34, 197, 94, 0.4));
-letter-spacing: -2px;
-}
-.subtitle {
-color: #c522af;
-text-transform: uppercase;
-letter-spacing: 6px;
-font-size: 0.9rem;
-text-shadow: 0 0 10px rgba(168, 85, 247, 0.5);
-}
-.flora-card {
-background: linear-gradient(145deg, rgba(13, 23, 15, 0.8), rgba(8, 12, 9, 0.95));
-border: 1px solid #14532d;
-border-left: 5px solid #22c55e;
-padding: 30px;
-margin: 25px 0;
-border-radius: 8px 20px 20px 8px;
-box-shadow: inset 0 -2px 10px rgba(34, 197, 94, 0.05), 0 10px 30px rgba(0,0,0,0.5);
-position: relative;
-z-index: 5;
-}
-.violet-card {
-border-left-color: #c6f755;
-border-color: #c6f755;
-background: linear-gradient(145deg, rgba(23, 10, 33, 0.6), rgba(12, 5, 18, 0.9));
-}
-.brown-card {
-border-left-color: #78350f;
-border-color: #451a03;
-background: linear-gradient(145deg, rgba(35, 18, 9, 0.7), rgba(15, 8, 4, 0.95));
-}
-.flora-card h3 { color: #22c55e; margin-top: 0; }
-.violet-card h3 { color: #c6f755; }
-.brown-card h3 { color: #d97706; }
-hr.vine {
-border: 0;
-height: 1px;
-background-image: linear-gradient(to right, rgba(0,0,0,0), #22c55e, rgba(0,0,0,0));
-margin: 20px 0;
-}
-hr.vine-violet {
-background-image: linear-gradient(to right, rgba(0,0,0,0), #c6f755, rgba(0,0,0,0));
-}
-.samplers-section { display: flex; gap: 30px; flex-wrap: wrap; margin-top: 40px; }
-.samplers-item { flex: 1; min-width: 280px; }
-.toxic-table {
-width: 100%;
-border-collapse: collapse;
-background: rgba(0,0,0,0.3);
-border-radius: 12px;
-overflow: hidden;
-border: 1px solid rgba(34, 197, 94, 0.15);
-}
-.toxic-table td, .toxic-table th {
-padding: 12px 15px;
-border-bottom: 1px solid rgba(34, 197, 94, 0.1);
-font-size: 0.9rem;
-}
-.toxic-table th { text-align: left; background: rgba(0,0,0,0.4); border-bottom: 2px solid #14532d; }
-.toxic-table tr:last-child td { border-bottom: none; }
-.toxic-table td:first-child { color: #94a3b8; font-weight: 500; }
-.toxic-table td:last-child { text-align: right; color: #22c55e; font-family: monospace; font-size: 1.1em;}
-.violet-table { border-color: rgba(168, 85, 247, 0.2); }
-.violet-table td, .violet-table th { border-bottom-color: rgba(168, 85, 247, 0.1); }
-.violet-table th { border-bottom-color: #c6f755; color: #c6f755; }
-.violet-table td:first-child { color: #22c55e; font-family: monospace; }
-.violet-table td:last-child { text-align: left; }
-.violet-table a { color: #f5e872; text-decoration: none; font-weight: bold; }
-.violet-table a:hover { color: #22c55e; }
-.violet-table tr:hover { background: rgba(168, 85, 247, 0.05); }
-.stat-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(160px, 1fr)); gap: 18px; margin: 30px 0; }
-.stat-item {
-text-align: center;
-padding: 18px;
-background: linear-gradient(180deg, #0a140d 0%, #050a06 100%);
-border: 1px solid #14532d;
-border-radius: 12px;
-box-shadow: inset 0 2px 10px rgba(34, 197, 94, 0.03), 0 8px 20px rgba(0,0,0,0.4);
-}
-.stat-value { display: block; font-size: 1.3rem; font-weight: bold; color: #22c55e; text-shadow: 0 0 10px rgba(34, 197, 94, 0.3); }
-.blume-credits ul { list-style: none; padding-left: 0; }
-.blume-credits li { margin-bottom: 12px; position: relative; padding-left: 25px; color: #cbd5e1; }
-.blume-credits li::before {
-content: '🌱';
-color: #c6f755;
-position: absolute;
-left: 0; top: 0;
-font-size: 0.9em;
-filter: drop-shadow(0 0 5px rgba(168, 85, 247, 0.6));
-}
-.blume-credits a { color: #f5e872; text-decoration: none; border-bottom: 1px dotted #f5e872; }
-.blume-credits a:hover { color: #22c55e; border-color: #22c55e; }
-</style>
+# Gemma-4-Giftige-Blume-31B-v1-W8A16-FP8
 
-<div class="blume-wrapper">
-<div style="font-size: 0.9rem; color: #94a3b8; margin-bottom: 15px; text-align: center;">by Nimbz</div>
-<div class="toxic-header">
-<img src="giftige-blume-31b-title.png" width="100%" style="border-radius: 15px; border: 1px solid #14532d; box-shadow: 0 15px 40px rgba(0,0,0,0.6); margin-bottom: 15px;">
-<h1 class="toxic-title"> 🌺 Giftige-Blume-31B-v1 🌻 </h1>
-<div class="subtitle">Beautiful in Prose • Rich in Variance</div>
-</div>
+Quantized version of [Blazed-Forge/Gemma-4-Giftige-Blume-31B-v1](https://huggingface.co/Blazed-Forge/Gemma-4-Giftige-Blume-31B-v1).
 
-<div class="stat-grid">
-<div class="stat-item">
-<span class="stat-value">AUTHENTIC</span>
-</div>
-<div class="stat-item">
-<span class="stat-value" style="color: #c6f755; text-shadow: 0 0 10px rgba(168, 85, 247, 0.4);">TOXIC</span>
-</div>
-<div class="stat-item">
-<span class="stat-value">CREATIVE</span>
-</div>
-</div>
+## Format
 
-<div class="flora-card">
-<h3>🌿 About The Model</h3>
-<div style="margin-bottom: 15px;">
+Offline-quantized **W8A16 FP8** in the
+[compressed-tensors](https://github.com/neuralmagic/compressed-tensors)
+`float-quantized` format: weights in `float8_e4m3fn` with per-output-channel
+symmetric scales, activations kept in bf16/fp16.
 
-With the model Giftige-Blume-31B, the idea was to create something that truly catches and portrays a character's persona genuinely. A character should act stubborn if they are stubborn, or refuse situations that *the character* doesn't seem fit. The Gemma-4-31B-it base and a lot of fine-tunes have a steep user adherence, which is usually great for assistant tasks, but not really fitting for immersive roleplay. </div>
-<hr class="vine">
-<div style="margin-bottom: 15px;">While Giftige-Blume-31B-v2 aims to be more logical and prompt adherent, Giftige-Blume-31B-v1 brings in a lot of mixed roleplay-centered models as it's knowledge core to provide rich and diverse roleplay scenarios, carrying over the genuine character portrayal, diverse prose and writing styles with each swipe.</div>
-<hr class="vine">
-<div>This model is made for a wide range of roleplay scenarios, with the goal to impersonate characters as they are described and not how the user wants them to be mid-chat. Expect uncensored roleplay without the deep-fried agreeability that is often associated with abliteration or similar techniques.</div>
-</div>
+## How it was quantized
 
-<div class="samplers-section">
-<div class="samplers-item">
-<h2 style="color: #cbd5e1; border-bottom: 1px solid #14532d; padding-bottom: 10px; margin-top: 0;">🧪 Samplers</h2>
-<table class="toxic-table">
-<tr><td>Temperature</td><td>0.8 - 1.2</td></tr>
-<tr><td>Top-K</td><td>0</td></tr>
-<tr><td>Top-P</td><td>0.95</td></tr>
-<tr><td>Min-P</td><td>0.1 - 0.2</td></tr>
-<tr><td>Adaptive-P Target</td><td style="color: #c6f755;">0.6</td></tr>
-<tr><td>Adaptive-P Decay</td><td style="color: #c6f755;">0.5</td></tr>
-</table>
-</div>
-<div class="samplers-item">
-<div class="flora-card violet-card" style="margin: 0;">
-<h3>🪻 GGUF Quants</h3>
-<div style="font-size: 0.9rem; color: #94a3b8; margin-bottom: 15px;">Select your vintage below.</div>
-<table class="toxic-table violet-table">
-<thead>
-<tr><th>Quant</th><th style="color: #cbd5e1;">Size</th><th>Link</th></tr>
-</thead>
-<tbody>
-<tr><td>Q4_K_S</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">17.8 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-GGUF/blob/main/Giftige-Blume-31B-v1-Q4_K_S.gguf">Download</a></td></tr>
-<tr><td>Q4_K_M</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">18.7 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-v1-GGUF/blob/main/Giftige-Blume-31B-v1-Q4_K_M.gguf">Download</a></td></tr>
-<tr><td>Q5_K_S</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">21.3 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-v1-GGUF/blob/main/Giftige-Blume-31B-v1-Q5_K_S.gguf">Download</a></td></tr>
-<tr><td>Q5_K_M</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">21.8 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-v1-GGUF/blob/main/Giftige-Blume-31B-v1-Q5_K_M.gguf">Download</a></td></tr>
-<tr><td>Q6_K</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">25.2 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-v1-GGUF/blob/main/Giftige-Blume-31B-v1-Q6_K.gguf">Download</a></td></tr>
-<tr><td>Q8_0</td><td style="color: #94a3b8; font-family: monospace; text-align: right;">32.6 GB</td><td><a href="https://huggingface.co/Nimbz/Gemma-4-Giftige-Blume-31B-v1-GGUF/blob/main/Giftige-Blume-31B-v1-Q8_0.gguf">Download</a></td></tr>
-</tbody>
-</table>
-</div>
-</div>
-</div>
- 
-<div class="flora-card violet-card" style="margin-top: 35px;">
-<h3>📜 Prompt Format</h3>
-<div style="margin-bottom: 15px;"><a href="https://huggingface.co/google/gemma-4-31B-it#2-thinking-mode-configuration" style="color: #f5e872; font-weight: bold; text-decoration: none; border-bottom: 1px dashed #f5e872;">Please refer to the original google/gemma-4-31b-it for the correct chat template.</a></div>
-<div style="color: #cbd5e1;">Let your frontend handle the chat template if possible (e.g., Chat Completion in SillyTavern).</div>
-<div style="background: rgba(0,0,0,0.5); padding: 15px; border-radius: 8px; border-left: 3px solid #c6f755; margin-top: 15px;">
-<strong style="color: #22c55e;">For Reasoning:</strong> Add <code>&lt;|think|&gt;</code> at the very beginning of the system prompt. Thinking happens between <code>&lt;|channel&gt;thought\n</code> and <code>&lt;channel|&gt;</code> tags.
-</div>
-</div>
+For each linear layer, every output row gets its own scale starting from
+`amax / 448`, refined by an MSE clip search over ~9 clip fractions
+(0.8–1.0× amax) picking the lowest-error scale per row. Weights are then
+quantized `q = e4m3(w / scale)` with round-to-nearest and saturation. This
+per-channel + clipping scheme gives better SNR than vLLM's online per-tensor
+`--quantization fp8` path.
 
-<div style="position: relative; z-index: 5; margin-top: 40px;">
-<h2 style="color: #cbd5e1; border-bottom: 1px solid #14532d; padding-bottom: 10px;">🧬 Merge Details</h2>
-<div style="font-size: 0.95rem; color: #94a3b8; line-height: 1.8; margin-bottom: 15px;">This model was merged in three distinct phases:</div>
-<div style="background: rgba(0,0,0,0.3); border: 1px dashed #451a03; padding: 15px; border-radius: 10px; color: #c6f755; text-align: left;">
+Only 2D linear projection weights are quantized (attention q/k/v/o, MLP
+gate/up/down). Embeddings, norms, lm_head, routers/experts, and the vision
+tower stay in bf16 and are listed in the checkpoint's `ignore` list, so vLLM
+leaves them untouched.
 
-<div class="flora-card">
-<h3>🏵️ 1. Phase</h3>
-<div style="margin-bottom: 15px;"><details>
-<summary>Broad RP Scenario: model_stock</summary>
+## Supported vLLM platforms
 
-```
-merge_method: model_stock
-base_model: ./LatitudeGames/G4-Equinox-31B
-tokenizer_source: base
-dtype: bfloat16
-parameters:
-normalize: true
-models:
-- model: ./LatitudeGames/G4-Equinox-31B
-- model: ./MRockatansky/G4-Gemma-4-31B-storymaxxed
-- model: ./MRockatansky/G4-Gemma-4-31B-storymaxxed2
-- model: ./AuriAetherwiing/G4-31B-Musica-v1
-```
-
-</details></p>
-</div>
-</div>
-<div class="flora-card">
-<h3>💐 2. Phase</h3>
-<div style="margin-bottom: 15px;"><details>
-<summary>Heretic RP Engine: breadcrumbs_ties</summary>
-
-```
-merge_method: breadcrumbs_ties
-base_model: ./Broad-RP-Scenario-model_stock_phase1
-tokenizer_source: base
-dtype: bfloat16
-
-modules:
-  text_decoder:
-    
-    slices:
-      - sources:
-          - model: ./Broad-RP-Scenario-model_stock_phase1
-            layer_range: [0, 12]
-            parameters:
-              weight: 0.90
-          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-            layer_range: [0, 12]
-            parameters:
-              density: 0.35
-              weight: 0.30            
-
-      - sources:
-          - model: ./Broad-RP-Scenario-model_stock_phase1
-            layer_range: [12, 24]
-            parameters:
-              weight: 0.85
-          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-            layer_range: [12, 24]
-            parameters:
-              density: 0.15
-              weight: 0.15
-          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
-            layer_range: [12, 24]
-            parameters:
-              density: 0.50
-              weight: 0.60
-
-      - sources:
-          - model: ./Broad-RP-Scenario-model_stock_phase1
-            layer_range: [24, 42]
-            parameters:
-              weight: 0.80
-          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-            layer_range: [24, 42]
-            parameters:
-              density: 0.30
-              weight: 0.35
-          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
-            layer_range: [24, 42]
-            parameters:
-              density: 0.70
-              weight: 0.80
-
-      - sources:
-          - model: ./Broad-RP-Scenario-model_stock_phase1
-            layer_range: [42, 54]
-            parameters:
-              weight: 0.80
-          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-            layer_range: [42, 54]
-            parameters:
-              density: 0.65
-              weight: 0.70
-          - model: /media/nimbz/cold_storage/LLM/WEIGHTS/G4-Mero-Artemis-31B-v0.3.1
-            layer_range: [42, 54]
-            parameters:
-              density: 0.40
-              weight: 0.45
-
-      - sources:
-          - model: ./Broad-RP-Scenario-model_stock_phase1
-            layer_range: [54, 60]
-            parameters:
-              weight: 0.85
-          - model: ./llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic
-            layer_range: [54, 60]
-            parameters:
-              density: 0.75
-              weight: 0.85     
-
-  vision_tower:
-    models:
-      - model: ./Broad-RP-Scenario-model_stock_phase1
-```
-
-</details></p>
-</div>
-</div>
-<div class="flora-card">
-<h3>🌳 3. Phase</h3>
-<div style="margin-bottom: 15px;"><details>
-<summary>Logical RP Melting: della_linear</summary>
-
-```
-merge_method: della_linear
-base_model: ./google/gemma-4-31B-it
-tokenizer_source: base
-dtype: bfloat16
-parameters:
-  epsilon: 0.05
-  lambda: 1.0
-  normalize: true
-  int8_mask: true
-
-modules:
-  text_decoder:
-    slices:
-      - sources:
-          - model: ./google/gemma-4-31B-it
-            layer_range: [0, 12]
-            parameters:
-              weight: 0.75
-          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
-            layer_range: [0, 12]
-            parameters:
-              density: 0.15
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.20
-          - model: ./Jackrong/Gemopus-4-31B-it
-            layer_range: [0, 12]
-            parameters:
-              density: 0.25
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.25
-          - model: ./Heretic-RP-Engine_breadcrumbs_ties
-            layer_range: [0, 12]
-            parameters:
-              density: 0.40
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.50
-
-      - sources:
-          - model: ./google/gemma-4-31B-it
-            layer_range: [12, 24]
-            parameters:
-              weight: 0.75
-          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
-            layer_range: [12, 24]
-            parameters:
-              density: 0.20
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.25
-          - model: ./Jackrong/Gemopus-4-31B-it
-            layer_range: [12, 24]
-            parameters:
-              density: 0.35
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.50
-          - model: ./Heretic-RP-Engine_breadcrumbs_ties
-            layer_range: [12, 24]
-            parameters:
-              density: 0.50
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.60
-
-      - sources:
-          - model: ./google/gemma-4-31B-it
-            layer_range: [24, 42]
-            parameters:
-              weight: 0.75 
-          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
-            layer_range: [24, 42]
-            parameters:
-              density: 0.30 
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.20
-          - model: ./Jackrong/Gemopus-4-31B-it
-            layer_range: [24, 42]
-            parameters:
-              density: 0.65
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.65 
-          - model: ./Heretic-RP-Engine_breadcrumbs_ties
-            layer_range: [24, 42]
-            parameters:
-              density: 0.85 
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.75  
-
-      - sources:
-          - model: ./google/gemma-4-31B-it
-            layer_range: [42, 54]
-            parameters:
-              weight: 0.75
-          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
-            layer_range: [42, 54]
-            parameters:
-              density: 0.25
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.45
-          - model: ./Heretic-RP-Engine_breadcrumbs_ties
-            layer_range: [42, 54]
-            parameters:
-              density: 0.40
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.75
-
-      - sources:
-          - model: ./google/gemma-4-31B-it
-            layer_range: [54, 60]
-            parameters:
-              weight: 0.70 
-          - model: ./p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B
-            layer_range: [54, 60]
-            parameters:
-              density: 0.25
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.50
-          - model: ./Heretic-RP-Engine_breadcrumbs_ties
-            layer_range: [54, 60]
-            parameters:
-              density: 0.35
-              weight:
-                - filter: embed_tokens
-                  value: 0.0
-                - filter: lm_head
-                  value: 0.0
-                - value: 0.65
-
-  vision_tower:
-    models:
-      - model: ./google/gemma-4-31B-it
-```
-
-</details></p>
-</div>
-</div>
-</div>
-<div style="position: relative; z-index: 5; margin-top: 40px;">
-<h2 style="color: #c6f755; border-bottom: 1px solid #14532d; padding-bottom: 10px;">🌸 Credits & Honors</h2>
-<div class="flora-card brown-card blume-credits" style="margin-top: 40px;">
-<ul>
-<li><b style="color: #22c55e;">Google DeepMind & The Open-Source Community:</b> For providing the brilliant base model, merges and fine-tunes that made this merge possible.</li>
-<li><b style="color: #22c55e;"><a href="https://www.reddit.com/r/BeaverAI/">The BeaverAI Community</a>:</b> Great people, always supportive, a really nice place to hang around.</li>
-<li><b style="color: #22c55e;">Mergekit:</b> Thank you arcee-ai for the mergekit, thank you zerofata for actually providing Gemma-4 support!</li>
-<li><b style="color: #22c55e;"><a href="https://huggingface.co/Ateron">Ateron</a>:</b> This guy.</li>
-<li><b style="color: #22c55e;">Google Gemini:</b> For hallucinating when I needed it the most.</li>
-</ul>
-</div>
+- **Intel XPU** — `XPUW8A16FP8LinearKernel` (the intended target).
+- **NVIDIA CUDA** (SM75+, i.e. Turing and newer) —
+  `HummingFP8ScaledMMLinearKernel` when the `humming` package is installed,
+  otherwise `MarlinFP8ScaledMMLinearKernel`.
+- **Not supported**: ROCm, CPU, TPU — vLLM has no W8A16-FP8 kernel for these
+  backends yet, so loading will fail with a "no kernel" error.