fix(card): substitute repo placeholders + drop vision note for text models

README.md

@@ -26,11 +26,10 @@ shims; vLLM auto-detects the quantization config from the embedded
 `config.json` at load time.
 
 This release is part of an ongoing series of vLLM-friendly quantized
-packs maintained by
+packs maintained by the [**atlas**](https://github.com/) self-evolving
+agent project, run by
 [**Alex Adamopoulos**](https://huggingface.co/aleada) at
-[**assert.gr**](https://assert.gr) — focused on filling gaps in the
-ecosystem where the most popular base models lack production-ready
-compressed-tensors weights.
+[**assert.gr**](https://assert.gr).
 
 ## Reproducibility
 
@@ -41,12 +40,18 @@ compressed-tensors weights.
 | Quantization recipe | GPTQModifier |
 | `scheme` | `W4A16` |
 | `targets` | `Linear` |
-| `ignore` | — |
+| `ignore` | `re:.*lm_head`, `re:.*vision_tower.*`, `re:.*multi_modal_projector.*` |
 | `sequential_targets` | — |
 | Calibration dataset | `lmms-lab/flickr30k` |
 | Calibration samples | 512 |
 | `max_seq_length` | 2048 |
-| Quantized size | 8.56 GiB |
+| Quantized size | 8.55 GiB |
+| Quantization time | 0.0 min (dual RTX 3090) |
+
+The pipeline that produced this artifact lives at
+[`tools/quantize/`](https://github.com/) in the atlas repository — see
+the README there for the full Docker build + run sequence and the
+per-model env-var recipes.
 
 ## License
 
@@ -61,20 +66,36 @@ docker run --runtime=nvidia --gpus all \
     -p 8000:8000 \
     -e HF_TOKEN=hf_XXX \
     vllm/vllm-openai:latest \
-    --model aleada/{REPO_NAME} \
+    --model aleada/Pixtral-12B-W4A16 \
+    --max-model-len 8192 \
+    --limit-mm-per-prompt 'image=1' \
     --gpu-memory-utilization 0.92 \
     --enable-prefix-caching
 ```
 
-vLLM auto-detects `compressed-tensors` from the model's `config.json`
-— passing `--quantization compressed-tensors` is allowed but redundant.
+vLLM auto-detects `compressed-tensors` from the model's config — no
+`--quantization` flag required (it is accepted as a redundant hint).
+Once vLLM is running, hit it with any OpenAI client:
+
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
+resp = client.chat.completions.create(
+    model="aleada/Pixtral-12B-W4A16",
+    messages=[{"role": "user", "content": "Hello"}],
+)
+print(resp.choices[0].message.content)
+```
 
 ## Hardware target
 
 Requires CUDA compute-capability ≥ 8.0 (Ampere or newer). Verified on
 **NVIDIA RTX 3090** (compute 8.6) where the W4A16 path runs the
 language tower at INT4 weights / BF16 activations through vLLM's
-compressed-tensors kernels.
+compressed-tensors kernels. Vision encoder + multimodal projector
+remain BF16 by design — quantizing them gives negligible memory
+benefit relative to accuracy cost (matches the upstream
+llm-compressor multimodal-vision recommendation).
 
 ## About the maintainer
 
@@ -82,8 +103,9 @@ Alex Adamopoulos is the founder of [assert.gr](https://assert.gr) and
 the engineer behind the **atlas** self-evolving AI agent platform.
 Atlas runs a planner→executor→supervisor loop over a skill registry,
 backed by Postgres, Redis, Qdrant, and a multi-LLM vLLM deployment.
-Quantization releases like this one keep the open-source ecosystem
-usable on consumer-grade hardware for self-hosted agent research.
+Quantization releases like this one keep the open-source VLM
+ecosystem usable on consumer-grade hardware for self-hosted agent
+research.
 
 Connect: