Add bench score 80/80/80

README.md

@@ -13,7 +13,7 @@ pipeline_tag: text-generation
 
 # Qwen3-8B-AWQ-INT4
 
-INT4 quantization of [`Qwen/Qwen3-8B`](https://huggingface.co/Qwen/Qwen3-8B). Built to run on a single 8 GB+ consumer GPU.
+INT4 quantization of [`Qwen/Qwen3-8B`](https://huggingface.co/Qwen/Qwen3-8B). Built to run on a single 12 GB+ consumer GPU.
 
 ## Footprint
 
@@ -23,7 +23,17 @@ INT4 quantization of [`Qwen/Qwen3-8B`](https://huggingface.co/Qwen/Qwen3-8B). Bu
 | Quantized weights | ~5.7 GB on disk |
 | Inference VRAM (incl. KV cache @ 32K context) | ~10 GB |
 
-Fits any 12 GB+ consumer card: RTX 3060 / 4060 / 4070 / 5070, even an integrated mobile GPU with shared memory. No homelab needed.
+Fits any 12 GB+ consumer card: RTX 3060 / 4060 / 4070 / 5070, even some integrated mobile GPUs with shared memory. No homelab needed.
+
+## Bench
+
+Scored on [`drawais/needle-1M-bench-mvp`](https://huggingface.co/datasets/drawais/needle-1M-bench-mvp) (50K-token haystack, real arxiv text):
+
+| Metric | Score |
+|---|---|
+| Overall recall | **80.0%** |
+| Paper-anchored | 80.0% |
+| Synthetic codes | 80.0% |
 
 ## Quick start
 
@@ -41,10 +51,6 @@ model = AutoModelForCausalLM.from_pretrained("drawais/Qwen3-8B-AWQ-INT4", device
 
 Native: 40,960 tokens. For longer contexts, enable YaRN rope-scaling per the base model's config.
 
-## Bench
-
-Leaderboard score on [`drawais/needle-1M-bench-mvp`](https://huggingface.co/datasets/drawais/needle-1M-bench-mvp) coming in next release.
-
 ## License
 
 Apache 2.0 (inherits from base model).