Hugging Face's logo

vasanth009
/
LC-lfm2.5-350m

Text Generation
MLX
Safetensors
English
lfm2
text-cleanup
dictation
course-correction
lora
conversational
5-bit
Model card Files
xet
 Community
vasanth009 commited on
Commit
76b33c3
·
verified ·
1 Parent(s): cef3c67

Fused LFM2.5-350M dictation cleanup + course-correction LoRA

Browse files
Files changed (10) hide show
  1. README.md +100 -0
  2. chat_template.jinja +7 -0
  3. config.json +73 -0
  4. generation_config.json +9 -0
  5. lora-adapter/adapter_config.json +41 -0
  6. lora-adapter/adapters.safetensors +3 -0
  7. model.safetensors +3 -0
  8. model.safetensors.index.json +342 -0
  9. tokenizer.json +0 -0
  10. tokenizer_config.json +22 -0
README.md ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: lfm-open
4
+ base_model: juanquivilla/sotto-cleanup-lfm25-350m-mlx-5bit
5
+ library_name: mlx
6
+ tags:
7
+ - mlx
8
+ - lfm2
9
+ - text-cleanup
10
+ - dictation
11
+ - course-correction
12
+ - lora
13
+ language:
14
+ - en
15
+ pipeline_tag: text-generation
16
+ ---
17
+
18
+ # LC-lfm2.5-350m — dictation cleanup with course-correction
19
+
20
+ A small, fast **on-device voice-dictation cleanup** model: it turns messy spoken
21
+ transcripts into clean written text, and — unlike most cleanup models — it
22
+ **honors spoken self-corrections** ("book the 7pm flight *no wait* the 9pm one"
23
+ → "Book the 9pm flight.").
24
+
25
+ Built for [MacWispr](https://github.com/vasanthsreeram/macwispr). This repo ships
26
+ the **fused** model (LoRA baked in) so you can pull and run it directly; the
27
+ standalone LoRA adapter is under [`lora-adapter/`](./lora-adapter).
28
+
29
+ - **Base:** `juanquivilla/sotto-cleanup-lfm25-350m-mlx-5bit` (LFM2.5-350M, MLX 5-bit)
30
+ - **Method:** LoRA (8 layers, 600 iters, LR 5e-6), fused into the base
31
+ - **Runtime:** MLX (Apple Silicon) — also loadable in Swift via `mlx-swift-lm` (LFM2)
32
+
33
+ ## What it does
34
+
35
+ 1. Removes fillers and stutters ("um", "uh", "that that" → "that").
36
+ 2. **Honors self-corrections** — drops the retracted item, keeps the replacement,
37
+ and keeps the rest of the sentence.
38
+ 3. Writes numbers as digits ("three seventy-five" → "375").
39
+ 4. Fixes light grammar/punctuation/capitalization without summarizing.
40
+
41
+ ## Prompt format
42
+
43
+ Trained on a raw completion format (**not** a chat template):
44
+
45
+ ```
46
+ ### Input:
47
+ {raw dictation}
48
+
49
+ ### Output:
50
+ ```
51
+
52
+ ## Usage (mlx-lm)
53
+
54
+ ```python
55
+ from mlx_lm import load, generate
56
+ from mlx_lm.sample_utils import make_sampler
57
+
58
+ model, tok = load("vasanth009/LC-lfm2.5-350m")
59
+ raw = "set the oven to three fifty no wait three seventy five for the lasagna"
60
+ prompt = f"### Input:\n{raw}\n\n### Output:\n"
61
+ out = generate(model, tok, prompt=prompt, max_tokens=64, sampler=make_sampler(temp=0.0))
62
+ print(out.split("###")[0].strip())
63
+ # -> Set the oven to 375 for the lasagna.
64
+ ```
65
+
66
+ To apply the LoRA to the base yourself instead of using the fused weights:
67
+
68
+ ```bash
69
+ mlx_lm.generate --model juanquivilla/sotto-cleanup-lfm25-350m-mlx-5bit \
70
+ --adapter-path lora-adapter --prompt "### Input:\n...\n\n### Output:\n"
71
+ ```
72
+
73
+ ## Honest evaluation (leak-free held-out)
74
+
75
+ Graded by an LLM judge on a **94-item held-out set generated on topics disjoint
76
+ from training** (0/94 overlap with the training data — verified). This is a real
77
+ generalization test, not memorized phrases.
78
+
79
+ | Model | Course-correction | Light cleanup | Preserve (anti over-edit) |
80
+ |-------|------------------:|--------------:|--------------------------:|
81
+ | Base (Sotto LFM2.5-350M) | 10/16 | 12/12 | 6/8 |
82
+ | **This model (+LoRA)** | **13/16** | 12/12 | **7/8** |
83
+
84
+ Course-correction is the headline improvement (10→13/16). Light cleanup was
85
+ already strong (tie). Latency ~50–100 ms/utterance on Apple Silicon.
86
+
87
+ ### Limitations
88
+
89
+ - The base is **5-bit quantized**; rare token corruptions can occur
90
+ ("simmer" → "smear"). A higher-precision base would reduce this.
91
+ - 350M parameters — capable for cleanup, not a general assistant. It only cleans
92
+ text; it does not answer questions in the transcript.
93
+ - Occasionally over-shortens a long correction (drops a trailing clause).
94
+
95
+ ## Provenance
96
+
97
+ Training data (course-correction / light-cleanup / preserve pairs) was generated
98
+ and QC-filtered with an LLM on fresh topics, with a hard leakage gate against the
99
+ held-out eval. See the MacWispr repo's `bench/polish_finetune/` for the full,
100
+ reproducible pipeline.
chat_template.jinja ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {{- bos_token -}}{%- set system_prompt = "" -%}{%- set ns = namespace(system_prompt="") -%}{%- if messages[0]["role"] == "system" -%} {%- set ns.system_prompt = messages[0]["content"] -%} {%- set messages = messages[1:] -%}{%- endif -%}{%- if tools -%} {%- set ns.system_prompt = ns.system_prompt + ("
2
+ " if ns.system_prompt else "") + "List of tools: <|tool_list_start|>[" -%} {%- for tool in tools -%} {%- if tool is not string -%} {%- set tool = tool | tojson -%} {%- endif -%} {%- set ns.system_prompt = ns.system_prompt + tool -%} {%- if not loop.last -%} {%- set ns.system_prompt = ns.system_prompt + ", " -%} {%- endif -%} {%- endfor -%} {%- set ns.system_prompt = ns.system_prompt + "]<|tool_list_end|>" -%}{%- endif -%}{%- if ns.system_prompt -%} {{- "<|im_start|>system
3
+ " + ns.system_prompt + "<|im_end|>
4
+ " -}}{%- endif -%}{%- for message in messages -%} {{- "<|im_start|>" + message["role"] + "
5
+ " -}} {%- set content = message["content"] -%} {%- if content is not string -%} {%- set content = content | tojson -%} {%- endif -%} {%- if message["role"] == "tool" -%} {%- set content = "<|tool_response_start|>" + content + "<|tool_response_end|>" -%} {%- endif -%} {{- content + "<|im_end|>
6
+ " -}}{%- endfor -%}{%- if add_generation_prompt -%} {{- "<|im_start|>assistant
7
+ " -}}{%- endif -%}
config.json ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Lfm2ForCausalLM"
4
+ ],
5
+ "block_auto_adjust_ff_dim": true,
6
+ "block_dim": 1024,
7
+ "block_ff_dim": 6656,
8
+ "block_ffn_dim_multiplier": 1.0,
9
+ "block_mlp_init_scale": 1.0,
10
+ "block_multiple_of": 256,
11
+ "block_norm_eps": 1e-05,
12
+ "block_out_init_scale": 1.0,
13
+ "block_use_swiglu": true,
14
+ "block_use_xavier_init": true,
15
+ "bos_token_id": 1,
16
+ "conv_L_cache": 3,
17
+ "conv_bias": false,
18
+ "conv_dim": 1024,
19
+ "conv_use_xavier_init": true,
20
+ "dtype": "bfloat16",
21
+ "eos_token_id": [
22
+ 7
23
+ ],
24
+ "hidden_size": 1024,
25
+ "initializer_range": 0.02,
26
+ "intermediate_size": 6656,
27
+ "layer_types": [
28
+ "conv",
29
+ "conv",
30
+ "full_attention",
31
+ "conv",
32
+ "conv",
33
+ "full_attention",
34
+ "conv",
35
+ "conv",
36
+ "full_attention",
37
+ "conv",
38
+ "full_attention",
39
+ "conv",
40
+ "full_attention",
41
+ "conv",
42
+ "full_attention",
43
+ "conv"
44
+ ],
45
+ "max_position_embeddings": 128000,
46
+ "model_type": "lfm2",
47
+ "norm_eps": 1e-05,
48
+ "num_attention_heads": 16,
49
+ "num_heads": 16,
50
+ "num_hidden_layers": 16,
51
+ "num_key_value_heads": 8,
52
+ "pad_token_id": 0,
53
+ "quantization": {
54
+ "group_size": 64,
55
+ "bits": 5,
56
+ "mode": "affine"
57
+ },
58
+ "quantization_config": {
59
+ "group_size": 64,
60
+ "bits": 5,
61
+ "mode": "affine"
62
+ },
63
+ "rope_parameters": {
64
+ "rope_theta": 1000000.0,
65
+ "rope_type": "default"
66
+ },
67
+ "rope_theta": 1000000.0,
68
+ "tie_word_embeddings": true,
69
+ "transformers_version": "5.3.0",
70
+ "use_cache": false,
71
+ "use_pos_enc": true,
72
+ "vocab_size": 65536
73
+ }
generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": [
5
+ 7
6
+ ],
7
+ "pad_token_id": 0,
8
+ "transformers_version": "5.3.0"
9
+ }
lora-adapter/adapter_config.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adapter_path": "/Users/vas/Documents/macwispr/bench/polish_finetune/adapters/sotto-lc-ft-clean",
3
+ "batch_size": 2,
4
+ "clear_cache_threshold": 0,
5
+ "config": null,
6
+ "data": "/Users/vas/Documents/macwispr/bench/polish_finetune/data/sotto_ft",
7
+ "fine_tune_type": "lora",
8
+ "grad_accumulation_steps": 1,
9
+ "grad_checkpoint": false,
10
+ "iters": 600,
11
+ "learning_rate": 5e-06,
12
+ "lora_parameters": {
13
+ "rank": 8,
14
+ "dropout": 0.0,
15
+ "scale": 20.0
16
+ },
17
+ "lr_schedule": null,
18
+ "mask_prompt": false,
19
+ "max_seq_length": 512,
20
+ "model": "juanquivilla/sotto-cleanup-lfm25-350m-mlx-5bit",
21
+ "num_layers": 8,
22
+ "optimizer": "adam",
23
+ "optimizer_config": {
24
+ "adam": {},
25
+ "adamw": {},
26
+ "muon": {},
27
+ "sgd": {},
28
+ "adafactor": {}
29
+ },
30
+ "project_name": null,
31
+ "report_to": null,
32
+ "resume_adapter_file": null,
33
+ "save_every": 200,
34
+ "seed": 42,
35
+ "steps_per_eval": 200,
36
+ "steps_per_report": 100,
37
+ "test": false,
38
+ "test_batches": 500,
39
+ "train": true,
40
+ "val_batches": 25
41
+ }
lora-adapter/adapters.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd8534ab1fdcf685db569702947a11b28c0058e3470ffe357f01e0a371d32f14
3
+ size 6039553
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3fc41058e03e8297bd75528a2e5d83d0cbf21e3a4403eb7c52a49233ef7e9347
3
+ size 243830226
model.safetensors.index.json ADDED
@@ -0,0 +1,342 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 243793408,
4
+ "total_parameters": 354483968
5
+ },
6
+ "weight_map": {
7
+ "model.embed_tokens.biases": "model.safetensors",
8
+ "model.embed_tokens.scales": "model.safetensors",
9
+ "model.embed_tokens.weight": "model.safetensors",
10
+ "model.embedding_norm.weight": "model.safetensors",
11
+ "model.layers.0.conv.conv.weight": "model.safetensors",
12
+ "model.layers.0.conv.in_proj.biases": "model.safetensors",
13
+ "model.layers.0.conv.in_proj.scales": "model.safetensors",
14
+ "model.layers.0.conv.in_proj.weight": "model.safetensors",
15
+ "model.layers.0.conv.out_proj.biases": "model.safetensors",
16
+ "model.layers.0.conv.out_proj.scales": "model.safetensors",
17
+ "model.layers.0.conv.out_proj.weight": "model.safetensors",
18
+ "model.layers.0.feed_forward.w1.biases": "model.safetensors",
19
+ "model.layers.0.feed_forward.w1.scales": "model.safetensors",
20
+ "model.layers.0.feed_forward.w1.weight": "model.safetensors",
21
+ "model.layers.0.feed_forward.w2.biases": "model.safetensors",
22
+ "model.layers.0.feed_forward.w2.scales": "model.safetensors",
23
+ "model.layers.0.feed_forward.w2.weight": "model.safetensors",
24
+ "model.layers.0.feed_forward.w3.biases": "model.safetensors",
25
+ "model.layers.0.feed_forward.w3.scales": "model.safetensors",
26
+ "model.layers.0.feed_forward.w3.weight": "model.safetensors",
27
+ "model.layers.0.ffn_norm.weight": "model.safetensors",
28
+ "model.layers.0.operator_norm.weight": "model.safetensors",
29
+ "model.layers.1.conv.conv.weight": "model.safetensors",
30
+ "model.layers.1.conv.in_proj.biases": "model.safetensors",
31
+ "model.layers.1.conv.in_proj.scales": "model.safetensors",
32
+ "model.layers.1.conv.in_proj.weight": "model.safetensors",
33
+ "model.layers.1.conv.out_proj.biases": "model.safetensors",
34
+ "model.layers.1.conv.out_proj.scales": "model.safetensors",
35
+ "model.layers.1.conv.out_proj.weight": "model.safetensors",
36
+ "model.layers.1.feed_forward.w1.biases": "model.safetensors",
37
+ "model.layers.1.feed_forward.w1.scales": "model.safetensors",
38
+ "model.layers.1.feed_forward.w1.weight": "model.safetensors",
39
+ "model.layers.1.feed_forward.w2.biases": "model.safetensors",
40
+ "model.layers.1.feed_forward.w2.scales": "model.safetensors",
41
+ "model.layers.1.feed_forward.w2.weight": "model.safetensors",
42
+ "model.layers.1.feed_forward.w3.biases": "model.safetensors",
43
+ "model.layers.1.feed_forward.w3.scales": "model.safetensors",
44
+ "model.layers.1.feed_forward.w3.weight": "model.safetensors",
45
+ "model.layers.1.ffn_norm.weight": "model.safetensors",
46
+ "model.layers.1.operator_norm.weight": "model.safetensors",
47
+ "model.layers.10.feed_forward.w1.biases": "model.safetensors",
48
+ "model.layers.10.feed_forward.w1.scales": "model.safetensors",
49
+ "model.layers.10.feed_forward.w1.weight": "model.safetensors",
50
+ "model.layers.10.feed_forward.w2.biases": "model.safetensors",
51
+ "model.layers.10.feed_forward.w2.scales": "model.safetensors",
52
+ "model.layers.10.feed_forward.w2.weight": "model.safetensors",
53
+ "model.layers.10.feed_forward.w3.biases": "model.safetensors",
54
+ "model.layers.10.feed_forward.w3.scales": "model.safetensors",
55
+ "model.layers.10.feed_forward.w3.weight": "model.safetensors",
56
+ "model.layers.10.ffn_norm.weight": "model.safetensors",
57
+ "model.layers.10.operator_norm.weight": "model.safetensors",
58
+ "model.layers.10.self_attn.k_layernorm.weight": "model.safetensors",
59
+ "model.layers.10.self_attn.k_proj.biases": "model.safetensors",
60
+ "model.layers.10.self_attn.k_proj.scales": "model.safetensors",
61
+ "model.layers.10.self_attn.k_proj.weight": "model.safetensors",
62
+ "model.layers.10.self_attn.out_proj.biases": "model.safetensors",
63
+ "model.layers.10.self_attn.out_proj.scales": "model.safetensors",
64
+ "model.layers.10.self_attn.out_proj.weight": "model.safetensors",
65
+ "model.layers.10.self_attn.q_layernorm.weight": "model.safetensors",
66
+ "model.layers.10.self_attn.q_proj.biases": "model.safetensors",
67
+ "model.layers.10.self_attn.q_proj.scales": "model.safetensors",
68
+ "model.layers.10.self_attn.q_proj.weight": "model.safetensors",
69
+ "model.layers.10.self_attn.v_proj.biases": "model.safetensors",
70
+ "model.layers.10.self_attn.v_proj.scales": "model.safetensors",
71
+ "model.layers.10.self_attn.v_proj.weight": "model.safetensors",
72
+ "model.layers.11.conv.conv.weight": "model.safetensors",
73
+ "model.layers.11.conv.in_proj.biases": "model.safetensors",
74
+ "model.layers.11.conv.in_proj.scales": "model.safetensors",
75
+ "model.layers.11.conv.in_proj.weight": "model.safetensors",
76
+ "model.layers.11.conv.out_proj.biases": "model.safetensors",
77
+ "model.layers.11.conv.out_proj.scales": "model.safetensors",
78
+ "model.layers.11.conv.out_proj.weight": "model.safetensors",
79
+ "model.layers.11.feed_forward.w1.biases": "model.safetensors",
80
+ "model.layers.11.feed_forward.w1.scales": "model.safetensors",
81
+ "model.layers.11.feed_forward.w1.weight": "model.safetensors",
82
+ "model.layers.11.feed_forward.w2.biases": "model.safetensors",
83
+ "model.layers.11.feed_forward.w2.scales": "model.safetensors",
84
+ "model.layers.11.feed_forward.w2.weight": "model.safetensors",
85
+ "model.layers.11.feed_forward.w3.biases": "model.safetensors",
86
+ "model.layers.11.feed_forward.w3.scales": "model.safetensors",
87
+ "model.layers.11.feed_forward.w3.weight": "model.safetensors",
88
+ "model.layers.11.ffn_norm.weight": "model.safetensors",
89
+ "model.layers.11.operator_norm.weight": "model.safetensors",
90
+ "model.layers.12.feed_forward.w1.biases": "model.safetensors",
91
+ "model.layers.12.feed_forward.w1.scales": "model.safetensors",
92
+ "model.layers.12.feed_forward.w1.weight": "model.safetensors",
93
+ "model.layers.12.feed_forward.w2.biases": "model.safetensors",
94
+ "model.layers.12.feed_forward.w2.scales": "model.safetensors",
95
+ "model.layers.12.feed_forward.w2.weight": "model.safetensors",
96
+ "model.layers.12.feed_forward.w3.biases": "model.safetensors",
97
+ "model.layers.12.feed_forward.w3.scales": "model.safetensors",
98
+ "model.layers.12.feed_forward.w3.weight": "model.safetensors",
99
+ "model.layers.12.ffn_norm.weight": "model.safetensors",
100
+ "model.layers.12.operator_norm.weight": "model.safetensors",
101
+ "model.layers.12.self_attn.k_layernorm.weight": "model.safetensors",
102
+ "model.layers.12.self_attn.k_proj.biases": "model.safetensors",
103
+ "model.layers.12.self_attn.k_proj.scales": "model.safetensors",
104
+ "model.layers.12.self_attn.k_proj.weight": "model.safetensors",
105
+ "model.layers.12.self_attn.out_proj.biases": "model.safetensors",
106
+ "model.layers.12.self_attn.out_proj.scales": "model.safetensors",
107
+ "model.layers.12.self_attn.out_proj.weight": "model.safetensors",
108
+ "model.layers.12.self_attn.q_layernorm.weight": "model.safetensors",
109
+ "model.layers.12.self_attn.q_proj.biases": "model.safetensors",
110
+ "model.layers.12.self_attn.q_proj.scales": "model.safetensors",
111
+ "model.layers.12.self_attn.q_proj.weight": "model.safetensors",
112
+ "model.layers.12.self_attn.v_proj.biases": "model.safetensors",
113
+ "model.layers.12.self_attn.v_proj.scales": "model.safetensors",
114
+ "model.layers.12.self_attn.v_proj.weight": "model.safetensors",
115
+ "model.layers.13.conv.conv.weight": "model.safetensors",
116
+ "model.layers.13.conv.in_proj.biases": "model.safetensors",
117
+ "model.layers.13.conv.in_proj.scales": "model.safetensors",
118
+ "model.layers.13.conv.in_proj.weight": "model.safetensors",
119
+ "model.layers.13.conv.out_proj.biases": "model.safetensors",
120
+ "model.layers.13.conv.out_proj.scales": "model.safetensors",
121
+ "model.layers.13.conv.out_proj.weight": "model.safetensors",
122
+ "model.layers.13.feed_forward.w1.biases": "model.safetensors",
123
+ "model.layers.13.feed_forward.w1.scales": "model.safetensors",
124
+ "model.layers.13.feed_forward.w1.weight": "model.safetensors",
125
+ "model.layers.13.feed_forward.w2.biases": "model.safetensors",
126
+ "model.layers.13.feed_forward.w2.scales": "model.safetensors",
127
+ "model.layers.13.feed_forward.w2.weight": "model.safetensors",
128
+ "model.layers.13.feed_forward.w3.biases": "model.safetensors",
129
+ "model.layers.13.feed_forward.w3.scales": "model.safetensors",
130
+ "model.layers.13.feed_forward.w3.weight": "model.safetensors",
131
+ "model.layers.13.ffn_norm.weight": "model.safetensors",
132
+ "model.layers.13.operator_norm.weight": "model.safetensors",
133
+ "model.layers.14.feed_forward.w1.biases": "model.safetensors",
134
+ "model.layers.14.feed_forward.w1.scales": "model.safetensors",
135
+ "model.layers.14.feed_forward.w1.weight": "model.safetensors",
136
+ "model.layers.14.feed_forward.w2.biases": "model.safetensors",
137
+ "model.layers.14.feed_forward.w2.scales": "model.safetensors",
138
+ "model.layers.14.feed_forward.w2.weight": "model.safetensors",
139
+ "model.layers.14.feed_forward.w3.biases": "model.safetensors",
140
+ "model.layers.14.feed_forward.w3.scales": "model.safetensors",
141
+ "model.layers.14.feed_forward.w3.weight": "model.safetensors",
142
+ "model.layers.14.ffn_norm.weight": "model.safetensors",
143
+ "model.layers.14.operator_norm.weight": "model.safetensors",
144
+ "model.layers.14.self_attn.k_layernorm.weight": "model.safetensors",
145
+ "model.layers.14.self_attn.k_proj.biases": "model.safetensors",
146
+ "model.layers.14.self_attn.k_proj.scales": "model.safetensors",
147
+ "model.layers.14.self_attn.k_proj.weight": "model.safetensors",
148
+ "model.layers.14.self_attn.out_proj.biases": "model.safetensors",
149
+ "model.layers.14.self_attn.out_proj.scales": "model.safetensors",
150
+ "model.layers.14.self_attn.out_proj.weight": "model.safetensors",
151
+ "model.layers.14.self_attn.q_layernorm.weight": "model.safetensors",
152
+ "model.layers.14.self_attn.q_proj.biases": "model.safetensors",
153
+ "model.layers.14.self_attn.q_proj.scales": "model.safetensors",
154
+ "model.layers.14.self_attn.q_proj.weight": "model.safetensors",
155
+ "model.layers.14.self_attn.v_proj.biases": "model.safetensors",
156
+ "model.layers.14.self_attn.v_proj.scales": "model.safetensors",
157
+ "model.layers.14.self_attn.v_proj.weight": "model.safetensors",
158
+ "model.layers.15.conv.conv.weight": "model.safetensors",
159
+ "model.layers.15.conv.in_proj.biases": "model.safetensors",
160
+ "model.layers.15.conv.in_proj.scales": "model.safetensors",
161
+ "model.layers.15.conv.in_proj.weight": "model.safetensors",
162
+ "model.layers.15.conv.out_proj.biases": "model.safetensors",
163
+ "model.layers.15.conv.out_proj.scales": "model.safetensors",
164
+ "model.layers.15.conv.out_proj.weight": "model.safetensors",
165
+ "model.layers.15.feed_forward.w1.biases": "model.safetensors",
166
+ "model.layers.15.feed_forward.w1.scales": "model.safetensors",
167
+ "model.layers.15.feed_forward.w1.weight": "model.safetensors",
168
+ "model.layers.15.feed_forward.w2.biases": "model.safetensors",
169
+ "model.layers.15.feed_forward.w2.scales": "model.safetensors",
170
+ "model.layers.15.feed_forward.w2.weight": "model.safetensors",
171
+ "model.layers.15.feed_forward.w3.biases": "model.safetensors",
172
+ "model.layers.15.feed_forward.w3.scales": "model.safetensors",
173
+ "model.layers.15.feed_forward.w3.weight": "model.safetensors",
174
+ "model.layers.15.ffn_norm.weight": "model.safetensors",
175
+ "model.layers.15.operator_norm.weight": "model.safetensors",
176
+ "model.layers.2.feed_forward.w1.biases": "model.safetensors",
177
+ "model.layers.2.feed_forward.w1.scales": "model.safetensors",
178
+ "model.layers.2.feed_forward.w1.weight": "model.safetensors",
179
+ "model.layers.2.feed_forward.w2.biases": "model.safetensors",
180
+ "model.layers.2.feed_forward.w2.scales": "model.safetensors",
181
+ "model.layers.2.feed_forward.w2.weight": "model.safetensors",
182
+ "model.layers.2.feed_forward.w3.biases": "model.safetensors",
183
+ "model.layers.2.feed_forward.w3.scales": "model.safetensors",
184
+ "model.layers.2.feed_forward.w3.weight": "model.safetensors",
185
+ "model.layers.2.ffn_norm.weight": "model.safetensors",
186
+ "model.layers.2.operator_norm.weight": "model.safetensors",
187
+ "model.layers.2.self_attn.k_layernorm.weight": "model.safetensors",
188
+ "model.layers.2.self_attn.k_proj.biases": "model.safetensors",
189
+ "model.layers.2.self_attn.k_proj.scales": "model.safetensors",
190
+ "model.layers.2.self_attn.k_proj.weight": "model.safetensors",
191
+ "model.layers.2.self_attn.out_proj.biases": "model.safetensors",
192
+ "model.layers.2.self_attn.out_proj.scales": "model.safetensors",
193
+ "model.layers.2.self_attn.out_proj.weight": "model.safetensors",
194
+ "model.layers.2.self_attn.q_layernorm.weight": "model.safetensors",
195
+ "model.layers.2.self_attn.q_proj.biases": "model.safetensors",
196
+ "model.layers.2.self_attn.q_proj.scales": "model.safetensors",
197
+ "model.layers.2.self_attn.q_proj.weight": "model.safetensors",
198
+ "model.layers.2.self_attn.v_proj.biases": "model.safetensors",
199
+ "model.layers.2.self_attn.v_proj.scales": "model.safetensors",
200
+ "model.layers.2.self_attn.v_proj.weight": "model.safetensors",
201
+ "model.layers.3.conv.conv.weight": "model.safetensors",
202
+ "model.layers.3.conv.in_proj.biases": "model.safetensors",
203
+ "model.layers.3.conv.in_proj.scales": "model.safetensors",
204
+ "model.layers.3.conv.in_proj.weight": "model.safetensors",
205
+ "model.layers.3.conv.out_proj.biases": "model.safetensors",
206
+ "model.layers.3.conv.out_proj.scales": "model.safetensors",
207
+ "model.layers.3.conv.out_proj.weight": "model.safetensors",
208
+ "model.layers.3.feed_forward.w1.biases": "model.safetensors",
209
+ "model.layers.3.feed_forward.w1.scales": "model.safetensors",
210
+ "model.layers.3.feed_forward.w1.weight": "model.safetensors",
211
+ "model.layers.3.feed_forward.w2.biases": "model.safetensors",
212
+ "model.layers.3.feed_forward.w2.scales": "model.safetensors",
213
+ "model.layers.3.feed_forward.w2.weight": "model.safetensors",
214
+ "model.layers.3.feed_forward.w3.biases": "model.safetensors",
215
+ "model.layers.3.feed_forward.w3.scales": "model.safetensors",
216
+ "model.layers.3.feed_forward.w3.weight": "model.safetensors",
217
+ "model.layers.3.ffn_norm.weight": "model.safetensors",
218
+ "model.layers.3.operator_norm.weight": "model.safetensors",
219
+ "model.layers.4.conv.conv.weight": "model.safetensors",
220
+ "model.layers.4.conv.in_proj.biases": "model.safetensors",
221
+ "model.layers.4.conv.in_proj.scales": "model.safetensors",
222
+ "model.layers.4.conv.in_proj.weight": "model.safetensors",
223
+ "model.layers.4.conv.out_proj.biases": "model.safetensors",
224
+ "model.layers.4.conv.out_proj.scales": "model.safetensors",
225
+ "model.layers.4.conv.out_proj.weight": "model.safetensors",
226
+ "model.layers.4.feed_forward.w1.biases": "model.safetensors",
227
+ "model.layers.4.feed_forward.w1.scales": "model.safetensors",
228
+ "model.layers.4.feed_forward.w1.weight": "model.safetensors",
229
+ "model.layers.4.feed_forward.w2.biases": "model.safetensors",
230
+ "model.layers.4.feed_forward.w2.scales": "model.safetensors",
231
+ "model.layers.4.feed_forward.w2.weight": "model.safetensors",
232
+ "model.layers.4.feed_forward.w3.biases": "model.safetensors",
233
+ "model.layers.4.feed_forward.w3.scales": "model.safetensors",
234
+ "model.layers.4.feed_forward.w3.weight": "model.safetensors",
235
+ "model.layers.4.ffn_norm.weight": "model.safetensors",
236
+ "model.layers.4.operator_norm.weight": "model.safetensors",
237
+ "model.layers.5.feed_forward.w1.biases": "model.safetensors",
238
+ "model.layers.5.feed_forward.w1.scales": "model.safetensors",
239
+ "model.layers.5.feed_forward.w1.weight": "model.safetensors",
240
+ "model.layers.5.feed_forward.w2.biases": "model.safetensors",
241
+ "model.layers.5.feed_forward.w2.scales": "model.safetensors",
242
+ "model.layers.5.feed_forward.w2.weight": "model.safetensors",
243
+ "model.layers.5.feed_forward.w3.biases": "model.safetensors",
244
+ "model.layers.5.feed_forward.w3.scales": "model.safetensors",
245
+ "model.layers.5.feed_forward.w3.weight": "model.safetensors",
246
+ "model.layers.5.ffn_norm.weight": "model.safetensors",
247
+ "model.layers.5.operator_norm.weight": "model.safetensors",
248
+ "model.layers.5.self_attn.k_layernorm.weight": "model.safetensors",
249
+ "model.layers.5.self_attn.k_proj.biases": "model.safetensors",
250
+ "model.layers.5.self_attn.k_proj.scales": "model.safetensors",
251
+ "model.layers.5.self_attn.k_proj.weight": "model.safetensors",
252
+ "model.layers.5.self_attn.out_proj.biases": "model.safetensors",
253
+ "model.layers.5.self_attn.out_proj.scales": "model.safetensors",
254
+ "model.layers.5.self_attn.out_proj.weight": "model.safetensors",
255
+ "model.layers.5.self_attn.q_layernorm.weight": "model.safetensors",
256
+ "model.layers.5.self_attn.q_proj.biases": "model.safetensors",
257
+ "model.layers.5.self_attn.q_proj.scales": "model.safetensors",
258
+ "model.layers.5.self_attn.q_proj.weight": "model.safetensors",
259
+ "model.layers.5.self_attn.v_proj.biases": "model.safetensors",
260
+ "model.layers.5.self_attn.v_proj.scales": "model.safetensors",
261
+ "model.layers.5.self_attn.v_proj.weight": "model.safetensors",
262
+ "model.layers.6.conv.conv.weight": "model.safetensors",
263
+ "model.layers.6.conv.in_proj.biases": "model.safetensors",
264
+ "model.layers.6.conv.in_proj.scales": "model.safetensors",
265
+ "model.layers.6.conv.in_proj.weight": "model.safetensors",
266
+ "model.layers.6.conv.out_proj.biases": "model.safetensors",
267
+ "model.layers.6.conv.out_proj.scales": "model.safetensors",
268
+ "model.layers.6.conv.out_proj.weight": "model.safetensors",
269
+ "model.layers.6.feed_forward.w1.biases": "model.safetensors",
270
+ "model.layers.6.feed_forward.w1.scales": "model.safetensors",
271
+ "model.layers.6.feed_forward.w1.weight": "model.safetensors",
272
+ "model.layers.6.feed_forward.w2.biases": "model.safetensors",
273
+ "model.layers.6.feed_forward.w2.scales": "model.safetensors",
274
+ "model.layers.6.feed_forward.w2.weight": "model.safetensors",
275
+ "model.layers.6.feed_forward.w3.biases": "model.safetensors",
276
+ "model.layers.6.feed_forward.w3.scales": "model.safetensors",
277
+ "model.layers.6.feed_forward.w3.weight": "model.safetensors",
278
+ "model.layers.6.ffn_norm.weight": "model.safetensors",
279
+ "model.layers.6.operator_norm.weight": "model.safetensors",
280
+ "model.layers.7.conv.conv.weight": "model.safetensors",
281
+ "model.layers.7.conv.in_proj.biases": "model.safetensors",
282
+ "model.layers.7.conv.in_proj.scales": "model.safetensors",
283
+ "model.layers.7.conv.in_proj.weight": "model.safetensors",
284
+ "model.layers.7.conv.out_proj.biases": "model.safetensors",
285
+ "model.layers.7.conv.out_proj.scales": "model.safetensors",
286
+ "model.layers.7.conv.out_proj.weight": "model.safetensors",
287
+ "model.layers.7.feed_forward.w1.biases": "model.safetensors",
288
+ "model.layers.7.feed_forward.w1.scales": "model.safetensors",
289
+ "model.layers.7.feed_forward.w1.weight": "model.safetensors",
290
+ "model.layers.7.feed_forward.w2.biases": "model.safetensors",
291
+ "model.layers.7.feed_forward.w2.scales": "model.safetensors",
292
+ "model.layers.7.feed_forward.w2.weight": "model.safetensors",
293
+ "model.layers.7.feed_forward.w3.biases": "model.safetensors",
294
+ "model.layers.7.feed_forward.w3.scales": "model.safetensors",
295
+ "model.layers.7.feed_forward.w3.weight": "model.safetensors",
296
+ "model.layers.7.ffn_norm.weight": "model.safetensors",
297
+ "model.layers.7.operator_norm.weight": "model.safetensors",
298
+ "model.layers.8.feed_forward.w1.biases": "model.safetensors",
299
+ "model.layers.8.feed_forward.w1.scales": "model.safetensors",
300
+ "model.layers.8.feed_forward.w1.weight": "model.safetensors",
301
+ "model.layers.8.feed_forward.w2.biases": "model.safetensors",
302
+ "model.layers.8.feed_forward.w2.scales": "model.safetensors",
303
+ "model.layers.8.feed_forward.w2.weight": "model.safetensors",
304
+ "model.layers.8.feed_forward.w3.biases": "model.safetensors",
305
+ "model.layers.8.feed_forward.w3.scales": "model.safetensors",
306
+ "model.layers.8.feed_forward.w3.weight": "model.safetensors",
307
+ "model.layers.8.ffn_norm.weight": "model.safetensors",
308
+ "model.layers.8.operator_norm.weight": "model.safetensors",
309
+ "model.layers.8.self_attn.k_layernorm.weight": "model.safetensors",
310
+ "model.layers.8.self_attn.k_proj.biases": "model.safetensors",
311
+ "model.layers.8.self_attn.k_proj.scales": "model.safetensors",
312
+ "model.layers.8.self_attn.k_proj.weight": "model.safetensors",
313
+ "model.layers.8.self_attn.out_proj.biases": "model.safetensors",
314
+ "model.layers.8.self_attn.out_proj.scales": "model.safetensors",
315
+ "model.layers.8.self_attn.out_proj.weight": "model.safetensors",
316
+ "model.layers.8.self_attn.q_layernorm.weight": "model.safetensors",
317
+ "model.layers.8.self_attn.q_proj.biases": "model.safetensors",
318
+ "model.layers.8.self_attn.q_proj.scales": "model.safetensors",
319
+ "model.layers.8.self_attn.q_proj.weight": "model.safetensors",
320
+ "model.layers.8.self_attn.v_proj.biases": "model.safetensors",
321
+ "model.layers.8.self_attn.v_proj.scales": "model.safetensors",
322
+ "model.layers.8.self_attn.v_proj.weight": "model.safetensors",
323
+ "model.layers.9.conv.conv.weight": "model.safetensors",
324
+ "model.layers.9.conv.in_proj.biases": "model.safetensors",
325
+ "model.layers.9.conv.in_proj.scales": "model.safetensors",
326
+ "model.layers.9.conv.in_proj.weight": "model.safetensors",
327
+ "model.layers.9.conv.out_proj.biases": "model.safetensors",
328
+ "model.layers.9.conv.out_proj.scales": "model.safetensors",
329
+ "model.layers.9.conv.out_proj.weight": "model.safetensors",
330
+ "model.layers.9.feed_forward.w1.biases": "model.safetensors",
331
+ "model.layers.9.feed_forward.w1.scales": "model.safetensors",
332
+ "model.layers.9.feed_forward.w1.weight": "model.safetensors",
333
+ "model.layers.9.feed_forward.w2.biases": "model.safetensors",
334
+ "model.layers.9.feed_forward.w2.scales": "model.safetensors",
335
+ "model.layers.9.feed_forward.w2.weight": "model.safetensors",
336
+ "model.layers.9.feed_forward.w3.biases": "model.safetensors",
337
+ "model.layers.9.feed_forward.w3.scales": "model.safetensors",
338
+ "model.layers.9.feed_forward.w3.weight": "model.safetensors",
339
+ "model.layers.9.ffn_norm.weight": "model.safetensors",
340
+ "model.layers.9.operator_norm.weight": "model.safetensors"
341
+ }
342
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<|startoftext|>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "eos_token": "<|im_end|>",
6
+ "extra_special_tokens": [],
7
+ "is_local": true,
8
+ "legacy": false,
9
+ "local_files_only": false,
10
+ "model_input_names": [
11
+ "input_ids",
12
+ "attention_mask"
13
+ ],
14
+ "model_max_length": 1000000000000000019884624838656,
15
+ "pad_token": "<|pad|>",
16
+ "sp_model_kwargs": {},
17
+ "spaces_between_special_tokens": false,
18
+ "tokenizer_class": "TokenizersBackend",
19
+ "tool_parser_type": "pythonic",
20
+ "use_default_system_prompt": false,
21
+ "use_fast": true
22
+ }