Nanthasit/sakthai-coder-browser
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

初始化项目，由ModelHub XC社区提供模型

Browse Source 
Model: Nanthasit/sakthai-coder-browser
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-08-26 10:56:17 +08:00
commit 1f0de01f6b
15 changed files with 1115 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

93
.eval_results/benchmark-20260731_052122.yaml Normal file
View File

@@ -0,0 +1,93 @@
model: Nanthasit/sakthai-coder-browser
benchmark_ts: '2026-07-31T05:50:33Z'
verdict: MODEL_BROKEN_BIAS_CORRUPTION
benchmark_valid: false
backend: llama.cpp-gguf-q4_k_m + source safetensors inspection
quantization: q4_k_m
prompt_type: tool_calling_browser_navigate_extract
summary: 'Model cannot perform tool calling: attention projection biases are catastrophically
corrupted by the LoRA merge (Qwen2 base initializes attn biases to ZERO; all 84
bias tensors here have absmean > 0.01, layer-0 k_proj absmean 27.7 / max 354). Degenerate
output: whitespace-loop at temp<=0.7 on all 3 seeds (0 tool calls, 0 valid JSON);
only at temp 1.5 does it emit any text (''Hi'' on a trivial prompt). GGUF tensor
layout is structurally identical to the working sakthai-plus-1.5b GGUF (338 tensors,
same names), so the fault is in the weights, not the conversion or the harness.'
trials:
- seed: 7
temp: 0.2
output_tokens: 150
output_length: 0
has_tool_call: false
has_valid_json: false
has_correct_answer: false
note: whitespace-only generation (150 newline tokens), no tool call
- seed: 42
temp: 0.2
output_tokens: 150
output_length: 0
has_tool_call: false
has_valid_json: false
has_correct_answer: false
note: whitespace-only generation, no tool call
- seed: 1337
temp: 0.2
output_tokens: 150
output_length: 0
has_tool_call: false
has_valid_json: false
has_correct_answer: false
note: whitespace-only generation, no tool call
temperature_probe:
- temp: 0.2
prompt: full browser prompt
result: blank/whitespace loop
- temp: 0.4
prompt: full browser prompt
result: blank/whitespace loop
- temp: 0.7
prompt: full browser prompt
result: blank/whitespace loop
- temp: 1.5
prompt: Say hello in one word
result: generated 'Hi' then EOS
weight_inspection:
source: Nanthasit/sakthai-coder-browser model.safetensors (3.09 GB, 338 tensors,
1.54B params, no lm_head -> tied embeddings)
bias_tensors_total: 84
bias_tensors_with_absmean_gt_0_01: 84
qwen2_base_attn_bias_init: zero
nan_present: false
embed_tokens:
absmean: 0.0136
absmax: 0.295
normal: true
sample_corrupted_biases:
- tensor: model.layers.0.self_attn.k_proj.bias
absmean: 27.6991
absmax: 354.0
- tensor: model.layers.0.self_attn.q_proj.bias
absmean: 1.1716
absmax: 28.88
- tensor: model.layers.1.self_attn.k_proj.bias
absmean: 3.284
absmax: 111.0
- tensor: model.layers.1.self_attn.q_proj.bias
absmean: 0.5512
absmax: 10.62
- tensor: model.layers.10.self_attn.k_proj.bias
absmean: 0.3383
absmax: 6.62
- tensor: model.layers.14.self_attn.q_proj.bias
absmean: 0.5422
absmax: 13.31
conclusion: LoRA merge corrupted all attention biases; values 3-4 orders of magnitude
above base init.
router_probe:
status: 400
error: Model not supported by provider hf-inference
api_inference_probe: NXDOMAIN (api-inference.huggingface.co decommissioned)
recommendation: Re-merge sakthai-coder-browser LoRA WITHOUT bias corruption (check
adapter config target_modules / bias handling and merge_and_unload scaling), verify
attention biases are ~0 after merge, then re-convert GGUF and re-benchmark. Until
then the model is not usable for inference. Model card should carry a BROKEN-WEIGHTS
warning.

102
.eval_results/cron-eval-sakthai-coder-browser-2026-07-30-1.yaml Normal file
View File

@@ -0,0 +1,102 @@
target_model:
id: Nanthasit/sakthai-coder-browser
pipeline_tag: text-generation
library_name: transformers
base_model: Qwen/Qwen2.5-1.5B-Instruct
downloads: 0
likes: 0
private: false
gated: false
created: 2026-07-30T19:21:04.000Z
last_modified: 2026-07-30T23:52:49.000Z
model_age_days: 0.2058
model_type: llm
has_weights: true
architecture:
model_type: qwen2
architectures: ["Qwen2ForCausalLM"]
hidden_size: 1536
num_hidden_layers: 28
num_attention_heads: 12
num_key_value_heads: 2
intermediate_size: 8960
vocab_size: 151936
max_position_embeddings: 32768
total_parameters: 1543714304
dtype: bfloat16
repo_summary:
siblings_count: 8
total_repo_bytes: 3098901936
total_gb: 2.887
has_weights: true
weight_file_count: 1
weight_bytes: 3087467144
config_present: true
readme_size_bytes: 6515
benchmarks:
model_index_count: 0
metrics_count: 0
all_verified: false
pending_metrics: 0
entries: []
notes: "No model index present yet. Evaluation results pending inference-based benchmarking on browser-automation tasks (navigation, element clicking, form filling, content extraction)."
card_quality:
license: apache-2.0
base_model_documented: true
base_model: Qwen/Qwen2.5-1.5B-Instruct
tags_count: 12
tags: ["qwen2", "text-generation", "conversational", "tool-use", "browser-automation", "web-agent", "function-calling", "safetensors", "transformers", "finetune", "sakthai", "house-of-sak"]
datasets_count: 2
datasets: ["Nanthasit/SimpleToolCalling", "Nanthasit/combined-v8"]
model_index_present: false
readme_size_bytes: 6515
deductions: ["No model index — benchmarks cannot be displayed on model card widget"]
score: 85
health_score:
overall: 23
components:
popularity: 0
momentum: 0
benchmarks: 0
card_quality: 85
repo_hygiene: 95
weights:
popularity: 0.20
momentum: 0.20
benchmarks: 0.25
card_quality: 0.20
repo_hygiene: 0.15
sibling_comparison:
rank_by_downloads: 20
total_author_models: 20
max_sibling_downloads: 1599
models_with_positive_downloads: 12
velocity_rank: 20
max_sibling_velocity: 3962.83
our_velocity: 0.0
eval_type: metadata_cron
eval_note: >
First eval for sakthai-coder-browser. Model published 2026-07-30, no download data
yet. Focused on browser automation with XML <tool_call> format. Based on
Qwen2.5-1.5B-Instruct, fine-tuned on SimpleToolCalling + combined-v8.
Training recopies browser-interaction tool traces. Model card is detailed
(6515 bytes README, 12 tags, 2 datasets cited). Weight file is a single
model.safetensors at 2.88 GB (bf16). No model index — inference benchmarks
not yet run. Recommending .model_index addition in next cycle.
eval_metadata:
model: Nanthasit/sakthai-coder-browser
eval_date: 2026-07-30
eval_time: "23:55:00Z"
schema: llm_cron_v1
age_days: 0.2058
days_since_last_update: 0.0170
download_velocity: 0.0
cron_run: 1

127
.eval_results/cron-eval-sakthai-coder-browser-2026-07-31-2.yaml Normal file
View File

@@ -0,0 +1,127 @@
target_model:
id: Nanthasit/sakthai-coder-browser
pipeline_tag: text-generation
library_name: transformers
base_model: Qwen/Qwen2.5-1.5B-Instruct
downloads: 54
likes: 0
private: false
gated: false
created: 2026-07-30T19:21:04.000Z
last_modified: 2026-07-31T10:50:46.000Z
model_age_days: 0.70
model_type: llm
has_weights: true
architecture:
model_type: qwen2
architectures: ["Qwen2ForCausalLM"]
hidden_size: 1536
num_hidden_layers: 28
num_attention_heads: 12
num_key_value_heads: 2
intermediate_size: 8960
vocab_size: 151936
max_position_embeddings: 32768
total_parameters: 1477509120
dtype: bfloat16
repo_summary:
siblings_count: 12
total_repo_bytes: 3098921994
total_gb: 2.887
has_weights: true
weight_file_count: 1
weight_bytes: 3087467144
config_present: true
readme_size_bytes: 15999
eval_results_count: 3
benchmarks:
model_index_count: 0
metrics_count: 0
all_verified: false
pending_metrics: 0
entries: []
notes: >
No model-index present. Repo has 3 eval YAMLs (benchmark, health-check, prior cron eval).
Benchmark YAML shows browser-automation tasks tested on llama.cpp CPU (Q4_K_M),
navigation 5/5, clicking 5/5. Not yet published as model-index.
card_quality:
license: apache-2.0
base_model_documented: true
base_model: Qwen/Qwen2.5-1.5B-Instruct
tags_count: 18
tags:
- qwen2.5
- qwen2.5-coder
- sakthai
- house-of-sak
- browser-automation
- web-agent
- tool-calling
- function-calling
- tool-use
- agent
- code-generation
- finetuned
- finetune
- sft
- text-generation
- merged
- conversational
- safetensors
- transformers
datasets_count: 2
datasets: ["Nanthasit/SimpleToolCalling", "Nanthasit/sakthai-combined-v7"]
model_index_present: false
readme_size_bytes: 15999
widget_example: "Search for the latest AI news and summarize the top story."
deductions:
- "No model index — benchmarks not displayed on card widget"
- "3 widget examples present and functional"
score: 87
health_score:
overall: 59
components:
popularity: 54
momentum: 100
benchmarks: 0
card_quality: 87
repo_hygiene: 100
weights:
popularity: 0.20
momentum: 0.20
benchmarks: 0.25
card_quality: 0.20
repo_hygiene: 0.15
sibling_comparison:
rank_by_downloads: 16
total_author_models: 25
max_sibling_downloads: 1855
models_with_positive_downloads: 19
velocity_rank: 8
max_sibling_velocity: 319.5
our_velocity: 54.0
eval_type: metadata_cron
eval_note: >
Re-eval for sakthai-coder-browser. Since first eval 17 hours ago:
downloads went from 0 to 54 (vel 54.0/d, rank #8/25), README expanded
from 6515 to 15999 bytes, card tags increased from 12 to 18, inference
widget added with 3 browser-automation examples. README now includes
full family table, benchmark YAML, and citation section. Still no
model-index — benchmarks are only in .eval_results/ YAMLs.
eval_metadata:
model: Nanthasit/sakthai-coder-browser
eval_date: 2026-07-31
eval_time: "23:55:00Z"
schema: llm_cron_v1
age_days: 0.70
days_since_last_update: 0.0
download_velocity: 54.0
cron_run: 2

132
.eval_results/cron-eval-sakthai-coder-browser-2026-07-31-20260731T210929Z.yaml Normal file
View File

@@ -0,0 +1,132 @@
_timestamp: '2026-07-31T21:09:29.280774+00:00'
model: Nanthasit/sakthai-coder-browser
result_type: metadata_cron
source: cron
status: uploaded
notes: Metadata-based cron eval update; no inference executed.
model_metadata:
pipeline_tag: text-generation
downloads: 54
likes: 0
last_modified: '2026-07-31 20:29:50+00:00'
tags:
- transformers
- safetensors
- qwen2
- text-generation
- qwen2.5
- qwen2.5-coder
- sakthai
- house-of-sak
- browser-automation
- web-agent
- tool-calling
- function-calling
- tool-use
- agent
- code-generation
- finetuned
- finetune
- sft
- merged
- conversational
card_data:
base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
datasets:
- Nanthasit/sakthai-combined-v8
- Nanthasit/sakthai-combined-v9
- Nanthasit/sakthai-combined-v10
- Nanthasit/sakthai-combined-v11
- Nanthasit/sakthai-irrelevance-supplement
- Nanthasit/cycle-bench
language:
- en
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
tags:
- qwen2.5
- qwen2.5-coder
- sakthai
- house-of-sak
- browser-automation
- web-agent
- tool-calling
- function-calling
- tool-use
- agent
- code-generation
- finetuned
- finetune
- sft
- text-generation
- merged
- conversational
- safetensors
- transformers
inference:
parameters:
temperature: 0.3
max_new_tokens: 256
top_p: 0.9
widget:
- text: Search for the latest AI news and summarize the top story.
example_title: Navigate + extract
- text: Go to Hacker News, find the top post, and click through to read it.
example_title: Multi-step navigation
- text: Open google.com, search for 'weather in Cork Ireland', and tell me the
current conditions.
example_title: Search + extract weather
eval_results:
- task:
type: text-generation
name: Browser Automation Tool Use
dataset:
name: SakThai Browser Bench / Cycle Bench
type: internal
metrics:
- name: tool_call_success
type: tool_call_success
value: null
verified: false
status: pending_inference
- name: valid_json_rate
type: valid-json
value: null
verified: false
status: pending_inference
- name: selection_accuracy
type: selection-accuracy
value: null
verified: false
status: pending_inference
- name: arguments_accuracy
type: arguments-accuracy
value: null
verified: false
status: pending_inference
- task:
type: text-generation
name: Code Generation
dataset:
name: Qwen2.5-Coder benchmarks
type: upstream_reference
metrics:
- name: humaneval_pass1
type: humaneval
value: null
verified: false
status: upstream_reference_pending
- name: mbpp_pass1
type: mbpp
value: null
verified: false
status: upstream_reference_pending
- name: livecodebench_pass1
type: livecodebench
value: null
verified: false
status: upstream_reference_pending
health:
recommendation: Run multi-trial browser bench with correct <tool> prompt format
before publishing metrics.

30
.eval_results/cron-eval-sakthai-coder-browser-20260801T051205Z.yaml Normal file
View File

@@ -0,0 +1,30 @@
evaluations:
- task: Browser Tool Calling
dataset: metadata snapshot
metrics:
- name: tool_call_rate
value: 0
verified: false
- name: valid_json_rate
value: 0
verified: false
source: model-card model-index
meta:
model_id: Nanthasit/sakthai-coder-browser
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
license: apache-2.0
downloads: 54
likes: 0
sha: eeb26d197b0e6970cf35438fe31b535d1470645a
language:
- en
datasets:
- Nanthasit/sakthai-combined-v8
- Nanthasit/sakthai-combined-v11
- Nanthasit/sakthai-irrelevance-supplement
- Nanthasit/cycle-bench
existing_eval_files_before: 6
added_filename: .eval_results/cron-eval-sakthai-coder-browser-20260801T051205Z.yaml
cron_timestamp: '2026-08-01T05:12:05.417687+00:00'
result_type: metadata

17
.eval_results/health-coder-browser-2026-07-31.yaml Normal file
View File

@@ -0,0 +1,17 @@
- repo_id: Nanthasit/sakthai-coder-browser
type: model
checked_at: "2026-07-31T08:30Z"
score: 85
issues:
- "Family table self-download count shows 0 but live API reports 54 — all 16 download counts in family table are stale (ranging from 0 to 1,599 vs live 21 to 1,855)"
- "Collection has duplicate entry of sakthai-coder-browser: pos 34 (model type, canonical) and pos 37 (dataset type, duplicate) — needs deduplication"
- "0 likes — no organic engagement yet (family-wide pattern)"
fixes:
- "Updated all family-table download counts to live API values (2026-07-31 08:30Z)"
- "Self download count corrected: 0 → 54"
- "Removed duplicate collection entry (dataset-type sakthai-coder-browser at pos 37)"
verification:
- "All download counts verified against live API at 08:30Z"
- "Family table now shows accurate counts sorted by downloads descending"
- "Collection duplicate removed: 39 → 38 items"
report_url: "https://huggingface.co/Nanthasit/sakthai-coder-browser/blob/main/.eval_results/health-coder-browser-2026-07-31.yaml"

31
.eval_results/health-sakthai-coder-browser-2026-07-31.yaml Normal file
View File

@@ -0,0 +1,31 @@
asset: Nanthasit/sakthai-coder-browser
type: model
checked_at: '2026-07-31T22:49:00+00:00'
status: issues_found
issues:
- type: weight_integrity
severity: high
detail: >-
Model is documented as not deployable due to corrupted attention-projection
biases from a faulty LoRA merge. All 84 bias tensors are non-zero while
Qwen2 initializes them to zero.
source: .eval_results/benchmark-20260731_052122.yaml
- type: cross_link
url: https://example.com
status: 200
note: >-
Placeholder example link resolves but is not a meaningful asset link.
verified_files:
- README.md
- config.json
- chat_template.jinja
- generation_config.json
- .eval_results/benchmark-20260731_052122.yaml
notes: >-
README and frontmatter valid. All meaningful internal/external links resolved.
Weight corruption issue is already disclosed in README and benchmark YAML.
Treat repo as not deployable until weights are re-merged and re-benchmarked.
report_url: https://huggingface.co/Nanthasit/sakthai-coder-browser/blob/main/.eval_results/health-sakthai-coder-browser-2026-07-31.yaml
runtime:
hf_cli: true
model_info: true