Update README: switch install to artemis-vlm v0.1.0 package

Drops the old 'pip install merlina; from src.artemis_vlm import ...' hack now that the ArtemisVLM model classes live in the dedicated Schneewolf-Labs/Artemis repo. Also drops the explicit model.all_tied_weights_keys = {} workaround — fixed in artemis-vlm v0.1.0 directly. Switches to AutoModelForCausalLM.from_pretrained() now that __init__.py registers with HF AutoConfig/AutoModel.

README.md

@@ -76,33 +76,36 @@ and it sets up a real Stage-1 run.
 
 ## What's next
 
-- **A3** — full Stage-1 (~1M samples on BLIP3o-Long-Caption) + Stage-2
-  multimodal instruction FFT with text-rehearsal so the underlying A2
-  text quality is retained.
-- **Artemis** — the polished named release after A3.
+- **A3** — full Stage-1 (~1M samples on BLIP3o-Long-Caption) currently training on
+  a single NVIDIA GB10. A3 is the projector-aligned successor to A3-preview.
+- **Artemis** — Stage-2 (multimodal instruction FFT with text rehearsal so A2's
+  reasoning / tool calling / identity survive). The named flagship multimodal
+  release after A3.
+
+## Install
+
+```bash
+pip install 'artemis-vlm @ git+https://github.com/Schneewolf-Labs/Artemis.git@v0.1.0'
+```
+
+The [`artemis-vlm`](https://github.com/Schneewolf-Labs/Artemis) package contains
+the model definition, processor, and data collator. On import, it registers
+`artemis_vlm` with HuggingFace AutoConfig and AutoModelForCausalLM so
+`from_pretrained()` resolves without `trust_remote_code`.
 
 ## Usage
 
 ```python
 import torch
-from transformers import AutoTokenizer
-# Requires the Schneewolf Labs Artemis VLM module:
-#   pip install merlina  # contains src.artemis_vlm
-# OR copy src/artemis_vlm.py from
-#   https://github.com/Schneewolf-Labs/Merlina
-from src.artemis_vlm import (
-    ArtemisVLMForConditionalGeneration,
-    ArtemisVLMProcessor,
-)
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import artemis_vlm  # registers ArtemisVLM with AutoConfig / AutoModel
 
-model = ArtemisVLMForConditionalGeneration.from_pretrained(
-    "schneewolflabs/A3-preview", dtype=torch.bfloat16
+model = AutoModelForCausalLM.from_pretrained(
+    "schneewolflabs/A3-preview", dtype=torch.bfloat16,
 ).to("cuda").eval()
-# transformers 5.x compat (untied weights — see Merlina #79 follow-up):
-model.all_tied_weights_keys = {}
 
 tok = AutoTokenizer.from_pretrained("schneewolflabs/A3-preview")
-processor = ArtemisVLMProcessor(
+processor = artemis_vlm.ArtemisVLMProcessor(
     tokenizer=tok, vision_config=model.visual.config,
     min_pixels=32 * 32, max_pixels=512 * 512,
 )