selfware 0.6.2

Your personal AI workshop — software you own, software that lasts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63

# Stress Test Configuration for Remote 122B SGLang Endpoint
# /media/thread/trebuchet6/qwen35/models/Qwen3.5-122B-A10B-NVFP4-yarn-1010k - Maxed Out Settings
#
# SGLang Config:
#   - Max Model Len: 1,010,000 tokens
#   - Context: 1010000 tokens

endpoint = "http://127.0.0.1:8000/v1"
model = "/media/thread/trebuchet6/qwen35/models/Qwen3.5-122B-A10B-NVFP4-yarn-1010k"
max_tokens = 8192
temperature = 0.6
context_length = 1010000

# MAXED OUT Concurrency Settings
[concurrency]
# Push to the limit (valid keys: max_streams, max_tools, max_global)
max_streams = 24
max_tools = 24
max_global = 48

[extra_body]
# Disable thinking for tool-heavy workflows
chat_template_kwargs = { enable_thinking = false }
top_p = 0.95
top_k = 20
min_p = 0.0

[safety]
allowed_paths = ["./**", "/tmp/**"]
denied_paths = ["**/.env", "**/secrets/**", "**/.ssh/**", "**/target/**"]
protected_branches = ["main"]

[agent]
streaming = false
max_iterations = 200
step_timeout_secs = 900
# XML-based tool calling for SGLang
native_function_calling = false
token_budget = 942000
token_safety_margin = 100000

[continuous_work]
enabled = true
checkpoint_interval_tools = 10
checkpoint_interval_secs = 300
auto_recovery = true
max_recovery_attempts = 6

[retry]
max_retries = 8
base_delay_ms = 1000
max_delay_ms = 70000

# Stress test specific settings
[stress_test]
# Number of concurrent agents to spawn
concurrent_agents = 16
# Duration of stress test in minutes
duration_minutes = 30
# Request rate (0 = unlimited)
requests_per_second = 0
# Enable token throughput measurement
measure_throughput = true