harn-vm 0.8.149

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
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159

# ---------- DeepInfra — open-weight OpenAI-compatible host -------------------
# DeepInfra serves open weights (DeepSeek, Qwen, Llama, Kimi, GPT-OSS) on a
# standard OpenAI chat-completions surface with native tool calls. Reasoning
# families expose an inline thinking trace; DeepSeek routes honor prompt
# caching. Catalog keys are `deepinfra/<hf-id>`, so patterns match the
# family substring.

[[provider.deepinfra]]
model_match = "*deepseek*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
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 = "inline"

[[provider.deepinfra]]
model_match = "*glm-5*"
native_tools = true
preferred_tool_format = "json"
tool_mode_parity = "native_unreliable"
tool_mode_parity_notes = "2026-06-20 Harn agent-loop smoke after parser fix: forced native/off emitted a malformed native call whose function name was the whole JSON payload; fenced JSON text-channel tools completed the loop."
structured_output = "native"
thinking_modes = ["enabled"]
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 = "inline"

[[provider.deepinfra]]
model_match = "*qwen3.7*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
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 = "inline"

# probed 2026-06-24 (docs/eval/provider-tool-mode-sweep-2026-06-24.md, N=5,
# forced-format single-tool authoring of a backslash-heavy Zig body): native
# 1/5 fidelity (bills empty completions), json 2/5 (flaky parse), text 5/5
# byte-clean. The provider-native channel cannot carry backslash-heavy code, so
# steer to the escape-free heredoc text channel.
[[provider.deepinfra]]
model_match = "*qwen3.6*"
native_tools = true
preferred_tool_format = "text"
tool_mode_parity = "native_unreliable"
tool_mode_parity_notes = "2026-06-24 forced-format sweep (N=5): DeepInfra Qwen3.6-35B-A3B native bills empty completions (1/5) and fenced-JSON is flaky (2/5); heredoc text carried a backslash-heavy Zig body byte-clean 5/5."
structured_output = "native"
thinking_modes = ["enabled"]
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.deepinfra]]
model_match = "*kimi-*"
native_tools = true
preferred_tool_format = "native"
structured_output = "native"
thinking_modes = ["enabled"]
prompt_caching = true
vision = true
vision_supported = true
video_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"

# DeepInfra-hosted GPT-OSS is the same Harmony model as cerebras/fireworks
# gpt-oss-120b. Placed BEFORE the catch-all `*` rule (first match wins) so it
# does not fall through to the non-reasoning default — without it
# `reasoning_required_for_tools` resolves OFF and the eval loop bills a
# noncommittal because gpt-oss calls tools INSIDE the chain-of-thought channel.
# Reasoning-effort thinking {low, medium, high}, `reasoning_required_for_tools =
# true`. Must NOT carry a Qwen-style `auto_reasoning_overrides = "off"`.
#
# TOOL CHANNEL — PINNED TO TEXT (matches the Fireworks #3505 gpt-oss precedent).
# DeepInfra's gpt-oss native channel is a Harmony footgun: a 2026-06-24 Harn
# agent-loop run (gpt-oss-120b, zig-feat, tool grounding present) saw DeepInfra
# native bill a non-empty completion (completion_tokens=86) with NO dispatchable
# tool call and NO answer — the action was serialized only in the private
# reasoning/commentary channel of the Harmony format, never emitted as a
# provider-native `tool_calls` entry. Repeated ~10x -> the run was unusable. This
# is the same class of defect reported across the open-source engines DeepInfra
# runs: vLLM #22578 (chat-completions tool_calls empty / missing arguments),
# vLLM #44216 (tool_choice="required" ignored, empty tool_calls + finish_reason
# stop), SGLang #8976 (commentary-channel crash) and #10738 (tool parser not
# working), openai/harmony #68 (tool intent stuck in commentary, never
# dispatched), and HuggingFace openai/gpt-oss-20b discussion #80 (function in
# reasoning_content with no tool_call structure). The clean+reliable channel on
# pay-per-token gpt-oss routes is TEXT (heredoc). `tool_mode_parity =
# "native_unreliable"` makes a `native` pin auto-correct to `text` via
# `validate_tool_format` with an explanatory `correction`, so no alias pin or
# `--tool-format native` can silently re-introduce the empty tool stream.
[[provider.deepinfra]]
model_match = "*gpt-oss*"
native_tools = false
preferred_tool_format = "text"
tool_mode_parity = "native_unreliable"
tool_mode_parity_notes = "2026-06-24 Harn agent-loop (gpt-oss-120b, zig-feat, tool grounding present): DeepInfra native billed completion_tokens=86 with no dispatchable tool call or answer (Harmony reasoning-channel-only / upstream contract violation), repeated ~10x -> run unusable. Text/heredoc is the clean pay-per-token channel. See vLLM #22578/#44216, SGLang #8976/#10738, openai/harmony #68."
structured_output = "native"
thinking_modes = ["effort"]
reasoning_effort_supported = true
reasoning_effort_levels = ["low", "medium", "high"]
reasoning_required_for_tools = 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.deepinfra]]
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"