sail / sail_scripts /kernel /test_kernel.py
muterornament's picture
Industrialize: Backup sovereign training pipeline and native kernel
0f6a233 verified
Raw
History Blame Contribute Delete
1.48 kB
import sail_kernel
print("sail_kernel version:", sail_kernel.__version__)
# Test 1: text cleaning
result = sail_kernel.clean_text("<b>Hello World</b>\n\n\n\nFoo")
print("clean_text:", repr(result))
# Test 2: batch cleaning
batch = sail_kernel.clean_batch(["<p>First</p>", "<b>Second text</b>", "Normal text"])
print("clean_batch:", batch)
# Test 3: dynamic token budget
budget = sail_kernel.compute_budget("Define photosynthesis. [2 marks]")
print("budget (2 marks):", budget)
budget8 = sail_kernel.compute_budget("Explain with examples how TCP/IP works. (8 marks)")
print("budget (8 marks):", budget8)
budget_deep = sail_kernel.compute_budget("Analyze and compare the causes of WW1 and WW2 in detail step-by-step")
print("budget (deep Q):", budget_deep)
# Test 4: reward scorer
score = sail_kernel.score_output("## Answer\n\n**Key point**: Use `cargo build`.\n\n- Item 1\n- Item 2", 500)
print("format-rich score:", score.format_score, "total:", score.total_score)
score_plain = sail_kernel.score_output("just plain text with no formatting at all", 100)
print("plain text score:", score_plain.format_score, "total:", score_plain.total_score)
# Test 5: batch reward scoring
scores = sail_kernel.score_batch([
"## Section\n\n**Bold**: some content\n\n- list item",
"plain text answer",
"## Deep Answer\n\n**Critical**: step by step analysis\n\n```python\ncode```"
], target_tokens=300)
print("batch scores:", scores)
print("\nAll SAIL kernel tests PASSED!")