thoddnn commited on
Commit
32aaf24
·
verified ·
1 Parent(s): 2d8edc0

Publish nano-planner-v1

Browse files
Files changed (5) hide show
  1. README.md +11 -0
  2. adapter_config.json +197 -0
  3. adapters.safetensors +3 -0
  4. manifest.json +22 -0
  5. prompt.txt +7 -0
README.md ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: thoddnn/Qwen3.5-2B-4bit
4
+ tags:
5
+ - mlx
6
+ - lora
7
+ ---
8
+
9
+ Fenn's Nano adapter for agent-search planning and fact extraction. Use with
10
+ the base revision and prompt specified in `manifest.json`. This adapter is
11
+ independent of the chat model and D3 verifier.
adapter_config.json ADDED
@@ -0,0 +1,197 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fine_tune_type": "lora",
3
+ "num_layers": 0,
4
+ "lora_parameters": {
5
+ "rank": 16,
6
+ "scale": 2.0,
7
+ "dropout": 0.05,
8
+ "keys": [
9
+ "language_model.model.layers.23.mlp.up_proj",
10
+ "language_model.model.layers.23.mlp.down_proj",
11
+ "language_model.model.layers.23.mlp.gate_proj",
12
+ "language_model.model.layers.23.self_attn.o_proj",
13
+ "language_model.model.layers.23.self_attn.v_proj",
14
+ "language_model.model.layers.23.self_attn.k_proj",
15
+ "language_model.model.layers.23.self_attn.q_proj",
16
+ "language_model.model.layers.22.mlp.up_proj",
17
+ "language_model.model.layers.22.mlp.down_proj",
18
+ "language_model.model.layers.22.mlp.gate_proj",
19
+ "language_model.model.layers.22.linear_attn.out_proj",
20
+ "language_model.model.layers.22.linear_attn.in_proj_a",
21
+ "language_model.model.layers.22.linear_attn.in_proj_b",
22
+ "language_model.model.layers.22.linear_attn.in_proj_z",
23
+ "language_model.model.layers.22.linear_attn.in_proj_qkv",
24
+ "language_model.model.layers.21.mlp.up_proj",
25
+ "language_model.model.layers.21.mlp.down_proj",
26
+ "language_model.model.layers.21.mlp.gate_proj",
27
+ "language_model.model.layers.21.linear_attn.out_proj",
28
+ "language_model.model.layers.21.linear_attn.in_proj_a",
29
+ "language_model.model.layers.21.linear_attn.in_proj_b",
30
+ "language_model.model.layers.21.linear_attn.in_proj_z",
31
+ "language_model.model.layers.21.linear_attn.in_proj_qkv",
32
+ "language_model.model.layers.20.mlp.up_proj",
33
+ "language_model.model.layers.20.mlp.down_proj",
34
+ "language_model.model.layers.20.mlp.gate_proj",
35
+ "language_model.model.layers.20.linear_attn.out_proj",
36
+ "language_model.model.layers.20.linear_attn.in_proj_a",
37
+ "language_model.model.layers.20.linear_attn.in_proj_b",
38
+ "language_model.model.layers.20.linear_attn.in_proj_z",
39
+ "language_model.model.layers.20.linear_attn.in_proj_qkv",
40
+ "language_model.model.layers.19.mlp.up_proj",
41
+ "language_model.model.layers.19.mlp.down_proj",
42
+ "language_model.model.layers.19.mlp.gate_proj",
43
+ "language_model.model.layers.19.self_attn.o_proj",
44
+ "language_model.model.layers.19.self_attn.v_proj",
45
+ "language_model.model.layers.19.self_attn.k_proj",
46
+ "language_model.model.layers.19.self_attn.q_proj",
47
+ "language_model.model.layers.18.mlp.up_proj",
48
+ "language_model.model.layers.18.mlp.down_proj",
49
+ "language_model.model.layers.18.mlp.gate_proj",
50
+ "language_model.model.layers.18.linear_attn.out_proj",
51
+ "language_model.model.layers.18.linear_attn.in_proj_a",
52
+ "language_model.model.layers.18.linear_attn.in_proj_b",
53
+ "language_model.model.layers.18.linear_attn.in_proj_z",
54
+ "language_model.model.layers.18.linear_attn.in_proj_qkv",
55
+ "language_model.model.layers.17.mlp.up_proj",
56
+ "language_model.model.layers.17.mlp.down_proj",
57
+ "language_model.model.layers.17.mlp.gate_proj",
58
+ "language_model.model.layers.17.linear_attn.out_proj",
59
+ "language_model.model.layers.17.linear_attn.in_proj_a",
60
+ "language_model.model.layers.17.linear_attn.in_proj_b",
61
+ "language_model.model.layers.17.linear_attn.in_proj_z",
62
+ "language_model.model.layers.17.linear_attn.in_proj_qkv",
63
+ "language_model.model.layers.16.mlp.up_proj",
64
+ "language_model.model.layers.16.mlp.down_proj",
65
+ "language_model.model.layers.16.mlp.gate_proj",
66
+ "language_model.model.layers.16.linear_attn.out_proj",
67
+ "language_model.model.layers.16.linear_attn.in_proj_a",
68
+ "language_model.model.layers.16.linear_attn.in_proj_b",
69
+ "language_model.model.layers.16.linear_attn.in_proj_z",
70
+ "language_model.model.layers.16.linear_attn.in_proj_qkv",
71
+ "language_model.model.layers.15.mlp.up_proj",
72
+ "language_model.model.layers.15.mlp.down_proj",
73
+ "language_model.model.layers.15.mlp.gate_proj",
74
+ "language_model.model.layers.15.self_attn.o_proj",
75
+ "language_model.model.layers.15.self_attn.v_proj",
76
+ "language_model.model.layers.15.self_attn.k_proj",
77
+ "language_model.model.layers.15.self_attn.q_proj",
78
+ "language_model.model.layers.14.mlp.up_proj",
79
+ "language_model.model.layers.14.mlp.down_proj",
80
+ "language_model.model.layers.14.mlp.gate_proj",
81
+ "language_model.model.layers.14.linear_attn.out_proj",
82
+ "language_model.model.layers.14.linear_attn.in_proj_a",
83
+ "language_model.model.layers.14.linear_attn.in_proj_b",
84
+ "language_model.model.layers.14.linear_attn.in_proj_z",
85
+ "language_model.model.layers.14.linear_attn.in_proj_qkv",
86
+ "language_model.model.layers.13.mlp.up_proj",
87
+ "language_model.model.layers.13.mlp.down_proj",
88
+ "language_model.model.layers.13.mlp.gate_proj",
89
+ "language_model.model.layers.13.linear_attn.out_proj",
90
+ "language_model.model.layers.13.linear_attn.in_proj_a",
91
+ "language_model.model.layers.13.linear_attn.in_proj_b",
92
+ "language_model.model.layers.13.linear_attn.in_proj_z",
93
+ "language_model.model.layers.13.linear_attn.in_proj_qkv",
94
+ "language_model.model.layers.12.mlp.up_proj",
95
+ "language_model.model.layers.12.mlp.down_proj",
96
+ "language_model.model.layers.12.mlp.gate_proj",
97
+ "language_model.model.layers.12.linear_attn.out_proj",
98
+ "language_model.model.layers.12.linear_attn.in_proj_a",
99
+ "language_model.model.layers.12.linear_attn.in_proj_b",
100
+ "language_model.model.layers.12.linear_attn.in_proj_z",
101
+ "language_model.model.layers.12.linear_attn.in_proj_qkv",
102
+ "language_model.model.layers.11.mlp.up_proj",
103
+ "language_model.model.layers.11.mlp.down_proj",
104
+ "language_model.model.layers.11.mlp.gate_proj",
105
+ "language_model.model.layers.11.self_attn.o_proj",
106
+ "language_model.model.layers.11.self_attn.v_proj",
107
+ "language_model.model.layers.11.self_attn.k_proj",
108
+ "language_model.model.layers.11.self_attn.q_proj",
109
+ "language_model.model.layers.10.mlp.up_proj",
110
+ "language_model.model.layers.10.mlp.down_proj",
111
+ "language_model.model.layers.10.mlp.gate_proj",
112
+ "language_model.model.layers.10.linear_attn.out_proj",
113
+ "language_model.model.layers.10.linear_attn.in_proj_a",
114
+ "language_model.model.layers.10.linear_attn.in_proj_b",
115
+ "language_model.model.layers.10.linear_attn.in_proj_z",
116
+ "language_model.model.layers.10.linear_attn.in_proj_qkv",
117
+ "language_model.model.layers.9.mlp.up_proj",
118
+ "language_model.model.layers.9.mlp.down_proj",
119
+ "language_model.model.layers.9.mlp.gate_proj",
120
+ "language_model.model.layers.9.linear_attn.out_proj",
121
+ "language_model.model.layers.9.linear_attn.in_proj_a",
122
+ "language_model.model.layers.9.linear_attn.in_proj_b",
123
+ "language_model.model.layers.9.linear_attn.in_proj_z",
124
+ "language_model.model.layers.9.linear_attn.in_proj_qkv",
125
+ "language_model.model.layers.8.mlp.up_proj",
126
+ "language_model.model.layers.8.mlp.down_proj",
127
+ "language_model.model.layers.8.mlp.gate_proj",
128
+ "language_model.model.layers.8.linear_attn.out_proj",
129
+ "language_model.model.layers.8.linear_attn.in_proj_a",
130
+ "language_model.model.layers.8.linear_attn.in_proj_b",
131
+ "language_model.model.layers.8.linear_attn.in_proj_z",
132
+ "language_model.model.layers.8.linear_attn.in_proj_qkv",
133
+ "language_model.model.layers.7.mlp.up_proj",
134
+ "language_model.model.layers.7.mlp.down_proj",
135
+ "language_model.model.layers.7.mlp.gate_proj",
136
+ "language_model.model.layers.7.self_attn.o_proj",
137
+ "language_model.model.layers.7.self_attn.v_proj",
138
+ "language_model.model.layers.7.self_attn.k_proj",
139
+ "language_model.model.layers.7.self_attn.q_proj",
140
+ "language_model.model.layers.6.mlp.up_proj",
141
+ "language_model.model.layers.6.mlp.down_proj",
142
+ "language_model.model.layers.6.mlp.gate_proj",
143
+ "language_model.model.layers.6.linear_attn.out_proj",
144
+ "language_model.model.layers.6.linear_attn.in_proj_a",
145
+ "language_model.model.layers.6.linear_attn.in_proj_b",
146
+ "language_model.model.layers.6.linear_attn.in_proj_z",
147
+ "language_model.model.layers.6.linear_attn.in_proj_qkv",
148
+ "language_model.model.layers.5.mlp.up_proj",
149
+ "language_model.model.layers.5.mlp.down_proj",
150
+ "language_model.model.layers.5.mlp.gate_proj",
151
+ "language_model.model.layers.5.linear_attn.out_proj",
152
+ "language_model.model.layers.5.linear_attn.in_proj_a",
153
+ "language_model.model.layers.5.linear_attn.in_proj_b",
154
+ "language_model.model.layers.5.linear_attn.in_proj_z",
155
+ "language_model.model.layers.5.linear_attn.in_proj_qkv",
156
+ "language_model.model.layers.4.mlp.up_proj",
157
+ "language_model.model.layers.4.mlp.down_proj",
158
+ "language_model.model.layers.4.mlp.gate_proj",
159
+ "language_model.model.layers.4.linear_attn.out_proj",
160
+ "language_model.model.layers.4.linear_attn.in_proj_a",
161
+ "language_model.model.layers.4.linear_attn.in_proj_b",
162
+ "language_model.model.layers.4.linear_attn.in_proj_z",
163
+ "language_model.model.layers.4.linear_attn.in_proj_qkv",
164
+ "language_model.model.layers.3.mlp.up_proj",
165
+ "language_model.model.layers.3.mlp.down_proj",
166
+ "language_model.model.layers.3.mlp.gate_proj",
167
+ "language_model.model.layers.3.self_attn.o_proj",
168
+ "language_model.model.layers.3.self_attn.v_proj",
169
+ "language_model.model.layers.3.self_attn.k_proj",
170
+ "language_model.model.layers.3.self_attn.q_proj",
171
+ "language_model.model.layers.2.mlp.up_proj",
172
+ "language_model.model.layers.2.mlp.down_proj",
173
+ "language_model.model.layers.2.mlp.gate_proj",
174
+ "language_model.model.layers.2.linear_attn.out_proj",
175
+ "language_model.model.layers.2.linear_attn.in_proj_a",
176
+ "language_model.model.layers.2.linear_attn.in_proj_b",
177
+ "language_model.model.layers.2.linear_attn.in_proj_z",
178
+ "language_model.model.layers.2.linear_attn.in_proj_qkv",
179
+ "language_model.model.layers.1.mlp.up_proj",
180
+ "language_model.model.layers.1.mlp.down_proj",
181
+ "language_model.model.layers.1.mlp.gate_proj",
182
+ "language_model.model.layers.1.linear_attn.out_proj",
183
+ "language_model.model.layers.1.linear_attn.in_proj_a",
184
+ "language_model.model.layers.1.linear_attn.in_proj_b",
185
+ "language_model.model.layers.1.linear_attn.in_proj_z",
186
+ "language_model.model.layers.1.linear_attn.in_proj_qkv",
187
+ "language_model.model.layers.0.mlp.up_proj",
188
+ "language_model.model.layers.0.mlp.down_proj",
189
+ "language_model.model.layers.0.mlp.gate_proj",
190
+ "language_model.model.layers.0.linear_attn.out_proj",
191
+ "language_model.model.layers.0.linear_attn.in_proj_a",
192
+ "language_model.model.layers.0.linear_attn.in_proj_b",
193
+ "language_model.model.layers.0.linear_attn.in_proj_z",
194
+ "language_model.model.layers.0.linear_attn.in_proj_qkv"
195
+ ]
196
+ }
197
+ }
adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3186a6baf3ee0230d740a1c54c11fec78f56ddede8f9774ade330c98dc690ce4
3
+ size 67323724
manifest.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "id": "nano-planner-v1",
3
+ "base_repo": "thoddnn/Qwen3.5-2B-4bit",
4
+ "base_revision": "da5c842efe38a8404c33e02bc9c59b2d81c99c34",
5
+ "schema": "nano-planner-v1",
6
+ "checkpoint_step": 50,
7
+ "adapter_dtype": "float32",
8
+ "files": {
9
+ "adapter_config.json": {
10
+ "sha256": "4116ad764b4a283dd3c418c5b3e3e3862b86e96dfcb6f69c27136890cd910766",
11
+ "bytes": 10880
12
+ },
13
+ "adapters.safetensors": {
14
+ "sha256": "3186a6baf3ee0230d740a1c54c11fec78f56ddede8f9774ade330c98dc690ce4",
15
+ "bytes": 67323724
16
+ },
17
+ "prompt.txt": {
18
+ "sha256": "84418edfc3c1af1a0785fc3c1ca2640741bca6e232e36695997f5da4a6c9d0b6",
19
+ "bytes": 2092
20
+ }
21
+ }
22
+ }
prompt.txt ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ Plan a local-file search. Return only JSON:
2
+ {"language":"en","file_types":[],"searches":[{"query":"retrieval terms","mode":"keyword"}],"facts":[{"question":"Complete question?","scope":"file","type":"boolean"}]}.
3
+ Use the user's language for searches/questions; preserve names and quoted text. Keys and enum values stay English. Extract ALL requirements without adding any.
4
+ file_types: only explicitly requested formats, chosen from image,video,pdf,doc,docx,txt,md,csv,xls,xlsx,ppt,pptx,eml. Otherwise []. Formats are metadata filters, never verification facts.
5
+ searches: 1-3 useful queries in execution order. Focus first, broaden next. semantic for visual subjects and conceptual meaning; keyword for document terms/names; exact for literal text; filename only for explicitly named files. For literal content, keyword AND exact searches with the same text can be useful. Never repeat the same query AND mode. Keep numeric comparisons and page/row constraints in facts, not keyword syntax. Meaningful identifiers, dates and visual counts may help retrieval. Search distinct visual subjects separately when useful. Automatic keyword/semantic fallback already exists.
6
+ facts: 1-8 questions, all must pass. boolean asks a question whose YES satisfies the requirement; omit target. Use boolean for comparisons, ranges, negation, identity, relations and substring containment. Exact values use year, count or value with a short string target: named year -> year; exact object count -> count; month/year or other exact value -> value. Do not generate answers. Preserve strict/inclusive comparisons, units, party roles and the correct date/amount field (payslip year means pay period). About a business does not imply issued by it.
7
+ scope=file for a document property or whole-image count. scope=any_unit for content on some page/row or a particular journey. Bind same-page/row/object/journey conditions in one question when separating would lose the relationship. Keep OR alternatives together. Do not invent a format, date, entity, requirement or answer. No stage field; no target for boolean facts.