harn-vm 0.8.154

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
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125

# ---------- Baseten Model APIs -----------------------------------------------
#
# Baseten serves current open-weight frontier models through an
# OpenAI-compatible /v1/chat/completions surface. The endpoint advertises
# tools/JSON/structured-output broadly, but live Harn probes on 2026-06-23
# found GLM-5.2 returns visible `<tool_call><arg_key>...` content instead of
# OpenAI `message.tool_calls` on the native channel. Harn's text-tool grammar
# did produce parseable calls, so pin GLM to text while keeping native tools for
# the Kimi, DeepSeek, GPT-OSS, and Nemotron routes that returned clean native
# calls in the same probe set.

[[provider.baseten]]
model_match = "*glm-5*"
native_tools = true
preferred_tool_format = "text"
tool_mode_parity = "native_unreliable"
tool_mode_parity_notes = "2026-06-23 live Baseten probe: forced native/off emitted visible `<tool_call><arg_key>...` content with no provider-native tool_calls; Harn heredoc text tools emitted parseable `<tool_call>\\nadd({ ... })\\n</tool_call>`."
structured_output = "native"
thinking_modes = ["enabled"]
auto_reasoning_overrides = { agent = "off", verify = "off", code = "off" }
prompt_caching = true
honors_chat_template_kwargs = true
chat_template_options_field = "chat_template_args"
reasoning_text_promotable = false
top_p_supported = true
top_k_supported = true
text_tool_wire_format_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.baseten]]
model_match = "*kimi-k2*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
honors_chat_template_kwargs = true
chat_template_options_field = "chat_template_args"
vision = true
vision_supported = true
text_tool_wire_format_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.baseten]]
model_match = "*deepseek-v4*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["effort"]
reasoning_effort_supported = true
reasoning_effort_levels = ["low", "medium", "high"]
prompt_caching = true
text_tool_wire_format_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 = "reasoning_summary"

[[provider.baseten]]
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"]
reasoning_required_for_tools = true
prompt_caching = true
text_tool_wire_format_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 = "reasoning_summary"

[[provider.baseten]]
model_match = "*nemotron*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
honors_chat_template_kwargs = true
chat_template_options_field = "chat_template_args"
top_p_supported = true
top_k_supported = true
text_tool_wire_format_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.baseten]]
model_match = "*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
text_tool_wire_format_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 = "none"