harn-vm 0.8.78

Async bytecode virtual machine for the Harn programming language
Documentation 
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
64
65
66
67
68

# ---------- Cerebras ----------------------------------------------------------

# Cerebras-hosted GPT-OSS accepts top-level `reasoning_effort` only for
# {low, medium, high}. It rejects `none` and `minimal` with
# wrong_api_format. Harn's provider-neutral `reasoning_policy: "off"` therefore
# floors to the lowest accepted effort (`low`) instead of sending a value the
# endpoint cannot template.
[[provider.cerebras]]
model_match = "gpt-oss-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["effort"]
reasoning_effort_supported = true
reasoning_effort_levels = ["low", "medium", "high"]
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "reasoning_summary"

# Cerebras GLM 4.7 is a public preview model with native tools, native
# structured output, and top-level `reasoning_effort`; `none` is the documented
# no-thinking value, while the older `disable_reasoning` knob is deprecated.
[[provider.cerebras]]
model_match = "zai-glm-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["effort"]
reasoning_effort_supported = true
reasoning_none_supported = true
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "inline"

[[provider.cerebras]]
model_match = "llama-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "none"

[[provider.cerebras]]
model_match = "qwen-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
prefers_xml_scaffolding = false
prefers_markdown_scaffolding = true
structured_output_mode = "native_json"
supports_assistant_prefill = false
prefers_role_developer = false
prefers_xml_tools = false
thinking_block_style = "none"