Jellyfish042 commited on
Commit
fef7b21
·
1 Parent(s): fcd1eb1
.claude/settings.local.json DELETED
@@ -1,17 +0,0 @@
1
- {
2
- "permissions": {
3
- "allow": [
4
- "Bash(git init:*)",
5
- "Bash(git add:*)",
6
- "Bash(git remote add:*)",
7
- "Bash(git push:*)",
8
- "Bash(git branch:*)",
9
- "Bash(git commit -m \"$\\(cat <<''EOF''\nFix Gradio compatibility for HuggingFace Spaces\n\n- Upgrade gradio to >=5.0.0 to fix API schema bug\n- Add server_name and server_port to demo.launch\\(\\)\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>\nEOF\n\\)\")",
10
- "Bash(git commit:*)",
11
- "Bash(git reset:*)",
12
- "Bash(and top-10 predictions\" to better reflect what users see in the tooltip.\nAlso updated color legend to match the swapped model positions.\n\nCo-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>\nEOF\n\\)\")",
13
- "Bash(git fetch:*)",
14
- "Bash(git pull:*)"
15
- ]
16
- }
17
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
.gitignore CHANGED
@@ -29,4 +29,5 @@ flagged/
29
  # Test artifacts
30
  tests/_out/
31
 
32
- .history/
 
 
29
  # Test artifacts
30
  tests/_out/
31
 
32
+ .history/
33
+ .claude/settings.local.json
precompute_example.py CHANGED
@@ -25,7 +25,10 @@ PRECOMPUTED_DIR = SCRIPT_DIR / "precomputed"
25
  QWEN_MODEL_ID = "Qwen/Qwen3-1.7B-Base"
26
  RWKV_MODEL_FILENAME = "rwkv7-g1j-1.5b-20260831-ctx16384.pth"
27
  RWKV_MODEL_DISPLAY_NAME = "RWKV7-G1J-1.5B"
28
- RWKV_MODEL_PATH = Path(r"E:\models\rwkv7-g1j-1.5b-20260831-ctx16384.pth")
 
 
 
29
 
30
  # Detect device
31
  # DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
@@ -37,7 +40,7 @@ def resolve_rwkv_model_path():
37
  """Resolve RWKV7 model path, preferring local absolute path."""
38
  from huggingface_hub import hf_hub_download
39
 
40
- if RWKV_MODEL_PATH.exists():
41
  return str(RWKV_MODEL_PATH)
42
 
43
  model_path = MODELS_DIR / RWKV_MODEL_FILENAME
@@ -115,7 +118,7 @@ def precompute_example():
115
  print(f"Example text length: {len(example_text)} characters")
116
 
117
  # Download and load models
118
- print(f"Resolving RWKV model path (prefer local: {RWKV_MODEL_PATH})...")
119
  rwkv_model_path = resolve_rwkv_model_path()
120
 
121
  print("Loading Qwen3-1.7B-Base...")
 
25
  QWEN_MODEL_ID = "Qwen/Qwen3-1.7B-Base"
26
  RWKV_MODEL_FILENAME = "rwkv7-g1j-1.5b-20260831-ctx16384.pth"
27
  RWKV_MODEL_DISPLAY_NAME = "RWKV7-G1J-1.5B"
28
+ # Optional local checkpoint path. Set RWKV_MODEL_PATH when running locally;
29
+ # leave it unset to download the checkpoint from Hugging Face.
30
+ _rwkv_model_path_env = os.getenv("RWKV_MODEL_PATH")
31
+ RWKV_MODEL_PATH = Path(_rwkv_model_path_env) if _rwkv_model_path_env else None
32
 
33
  # Detect device
34
  # DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
40
  """Resolve RWKV7 model path, preferring local absolute path."""
41
  from huggingface_hub import hf_hub_download
42
 
43
+ if RWKV_MODEL_PATH is not None and RWKV_MODEL_PATH.exists():
44
  return str(RWKV_MODEL_PATH)
45
 
46
  model_path = MODELS_DIR / RWKV_MODEL_FILENAME
 
118
  print(f"Example text length: {len(example_text)} characters")
119
 
120
  # Download and load models
121
+ print(f"Resolving RWKV model path (prefer local: {RWKV_MODEL_PATH or 'none'})...")
122
  rwkv_model_path = resolve_rwkv_model_path()
123
 
124
  print("Loading Qwen3-1.7B-Base...")
tests/golden/README.md CHANGED
@@ -9,13 +9,13 @@ Suggested naming:
9
  Keep these files in sync with `tests/samples/`.
10
 
11
  Generate snapshots with:
12
- `conda run -n torch2 python tests/generate_snapshots.py --out tests/golden`
13
 
14
  Generate a candidate snapshot:
15
- `conda run -n torch2 python tests/generate_snapshots.py --out tests/_out`
16
 
17
  Compare snapshots:
18
- `conda run -n torch2 python tests/compare_snapshots.py --baseline tests/golden/stress.render_model.json --candidate tests/_out/stress.render_model.json`
19
 
20
  Compare HTML output:
21
- `conda run -n torch2 python tests/compare_html.py --baseline tests/golden/stress.output.html --candidate tests/_out/stress.output.html`
 
9
  Keep these files in sync with `tests/samples/`.
10
 
11
  Generate snapshots with:
12
+ `python tests/generate_snapshots.py --out tests/golden`
13
 
14
  Generate a candidate snapshot:
15
+ `python tests/generate_snapshots.py --out tests/_out`
16
 
17
  Compare snapshots:
18
+ `python tests/compare_snapshots.py --baseline tests/golden/stress.render_model.json --candidate tests/_out/stress.render_model.json`
19
 
20
  Compare HTML output:
21
+ `python tests/compare_html.py --baseline tests/golden/stress.output.html --candidate tests/_out/stress.output.html`