main下最新版本Windows下编译运行不成功

[TangLaoya]

Windows server 2019离线环境，纯CPU。
使用附件所示脚本setup_clawcode_offline.ps1编译后，在D:\claw-code\rust\target\release下生成claw.exe
运行：
PS C:\AI_research\Quad_bipartite_label_AI> C:\claude_code\claw\clawcode\claw-code\rust\target\release\claw init
Init
Project C:\AI_research\Quad_bipartite_label_AI
.claw/ created
.claw.json created
.gitignore created
CLAUDE.md skipped (already exists)
Next step Review and tailor the generated guidance

PS C:\AI_research\Quad_bipartite_label_AI> C:\claude_code\claw\clawcode\claw-code\rust\target\release\claw
�[38;5;196m██████╗██╗ █████╗ ██╗ ██╗
██╔════╝██║ ██╔══██╗██║ ██║
██║ ██║ ███████║██║ █╗ ██║
██║ ██║ ██╔══██║██║███╗██║
╚██████╗███████╗██║ ██║╚███╔███╔╝
╚═════╝╚══════╝╚═╝ ╚═╝ ╚══╝╚══╝�[0m �[38;5;208mCode�[0m 🦞

�[2mModel�[0m Qwen3.6-35B-A3B-Instruct
�[2mPermissions�[0m danger-full-access
�[2mBranch�[0m unknown
�[2mWorkspace�[0m clean
�[2mDirectory�[0m C:\AI_research\Quad_bipartite_label_AI
�[2mSession�[0m session-1777451903343-0
�[2mAuto-save�[0m \?\C:\AI_research\Quad_bipartite_label_AI.claw\sessions\ff54a51b09285dd9\session-1777451903343-0.jsonl

Type �[1m/help�[0m for commands · �[1m/status�[0m for live context · �[2m/resume latest�[0m jumps back to the newest session · �[1m/diff�[0m then �[1m/commit�[0m to ship · �[2mTab�[0m for workflow completions · �[2mShift+Enter�[0m for newline
Connected: Qwen3.6-35B-A3B-Instruct via anthropic

请严格按照CLAUDE.md自动执行所有任务，执行完一条自动输出下一条，直到全部完成。
我来开始执行任务。首先读取记忆文件和任务清单。<tool_call>
<function=read_file>
<parameter=path>
C:\AI_research\Quad_bipartite_label_AI.claw\memory\skills.md

C:\AI_research\Quad_bipartite_label_AI.claw\memory\tasks.md C:\AI_research\Quad_bipartite_label_AI.claw\memory\modifycode.md ✔ ✨ Done

很快运行结束，并未按CLAUDE.md和..claw\memory\tasks.md
的要求完成任务。

本地通过llama-server启动LLM服务：
.\llama-server.exe -t 24 --mlock --no-mmap -ngl 0 -m .\Qwen3.6-35B-A3B\Qwen3.6-35B-A3B-Instruct-Pure-GGUF\Qwen3.6-35B-A3B-q8_0.gguf --jinja --chat-template-file .\Qwen3.6-35B-A3B\Qwen3.6-35B-A3B-Instruct-Pure-GGUF\qwen3_nonthinking.jinja --skip-chat-parsing --offline -fa on --temp 0.1 --top-p 0.95 --top-k 20 --min-p 0.01 --repeat-penalty 1.0 -c 262144 --host 0.0.0.0 --port 8080 --cache-type-k q8_0 --cache-type-v q8_0 --ubatch-size 2048 --batch-size 2048 --cache-prompt

此外，还设置了以下环境变量：
ANTHROPIC_BASE_URL=http://localhost:8080/
ANTHROPIC_API_KEY=local-llama-server-no-key-required
ANTHROPIC_MODEL=Qwen3.6-35B-A3B-Instruct
CLAW_TOOL_EXECUTION=true
RUST_LOG=info

请问还需要什么设置？或者是当前版本不支持这个模型，需要使用什么模型才可以工作？

谢谢！

附：
powershell脚本
<#
.SYNOPSIS
Offline build script for Claw Code (v17.6) - removed invalid dependencies.
.DESCRIPTION
Ensures the bundled MinGW is used for all C/C++ compilation. Detects mingw64 subfolder.
Configures .env for the specified llama-server model.
.NOTES
Version: 17.6
#>

$ErrorActionPreference = "Stop"
$ProgressPreference = "SilentlyContinue"

function Write-Step { param([string]$m) Write-Host $m -ForegroundColor Yellow }
function Write-Success { param([string]$m) Write-Host " $m" -ForegroundColor Green }
function Write-Warning { param([string]$m) Write-Host " WARNING: $m" -ForegroundColor Yellow }
function Write-ErrorExit { param([string]$m) Write-Host " ERROR: $m" -ForegroundColor Red; exit 1 }

$BaseDir = Join-Path $PSScriptRoot "clawcode"
$DepsDir = Join-Path $BaseDir "deps"
$SourceDir = Join-Path $BaseDir "claw-code"
$RustDir = Join-Path $DepsDir "rust"
$MinGWDir = Join-Path $DepsDir "mingw"

llama-server configuration

$LlamaserverHost = "localhost"
$LlamaserverPort = "8080"
$LlamaserverModel = "Qwen3-Coder-Next-80B"

Write-Host "========================================" -ForegroundColor Cyan
Write-Host "Claw Code - Offline Build Script v17.6" -ForegroundColor Cyan
Write-Host "========================================" -ForegroundColor Cyan
Write-Host ""

function Invoke-CargoCommand {
param(
[string]$Command,
[string]$Arguments,
[string]$WorkingDirectory,
[string]$LogPrefix
)
$logFile = Join-Path $env:TEMP "$LogPrefix`$([Guid]::NewGuid()).log"
$errFile = "$logFile.err"
$proc = Start-Process -FilePath $Command -ArgumentList $Arguments -WorkingDirectory $WorkingDirectory -NoNewWindow -Wait -PassThru -RedirectStandardOutput $logFile -RedirectStandardError $errFile
$success = ($proc.ExitCode -eq 0)
if (-not $success) {
Write-Host " --- Begin of error log (stdout) ---" -ForegroundColor Red
Get-Content $logFile -ErrorAction SilentlyContinue | ForEach-Object { Write-Host " $" -ForegroundColor DarkRed }
Write-Host " --- End of stdout ---" -ForegroundColor Red
if (Test-Path $errFile) {
$errContent = Get-Content $errFile -Raw
if ($errContent.Trim()) {
Write-Host " --- Begin of stderr ---" -ForegroundColor Red
Get-Content $errFile | ForEach-Object { Write-Host " $_" -ForegroundColor DarkRed }
Write-Host " --- End of stderr ---" -ForegroundColor Red
}
}
Write-Host " Full logs saved to: $logFile and $errFile" -ForegroundColor Yellow
} else {
Remove-Item $logFile, $errFile -ErrorAction SilentlyContinue
}
return $success
}

--- 0. Verify base structure ---

Write-Step "[0/7] Verifying base structure..."
if (!(Test-Path $BaseDir)) { Write-ErrorExit "Folder not found: $BaseDir" }
if (!(Test-Path $SourceDir)) { Write-ErrorExit "Source code not found: $SourceDir" }
if (!(Test-Path $RustDir)) { Write-ErrorExit "Rust toolchain not found: $RustDir" }
if (!(Test-Path $MinGWDir)) { Write-ErrorExit "MinGW directory not found at $MinGWDir." }
Write-Success "Structure OK."

--- 1. Locate Rust toolchain and MinGW ---

Write-Step "[1/7] Setting up Rust and MinGW environment..."
$toolchainsDir = Join-Path $RustDir "toolchains"
$rustcPath = $null
$cargoPath = $null
if (Test-Path $toolchainsDir) {
$found = Get-ChildItem $toolchainsDir -Directory | Where-Object { Test-Path (Join-Path $_.FullName "bin\rustc.exe") }
if ($found) {
$realBinDir = Join-Path $found[0].FullName "bin"
$rustcPath = Join-Path $realBinDir "rustc.exe"
$cargoPath = Join-Path $realBinDir "cargo.exe"
}
}
if (-not $rustcPath) { Write-ErrorExit "Valid rustc.exe not found." }

$env:CARGO_HOME = $RustDir
$env:RUSTUP_HOME = $RustDir

--- Locate MinGW bin directory ---

$MinGWBin = $null
$possibleBins = @(
(Join-Path $MinGWDir "bin"),
(Join-Path $MinGWDir "mingw64\bin")
)
foreach ($candidate in $possibleBins) {
if (Test-Path (Join-Path $candidate "gcc.exe")) {
$MinGWBin = $candidate
break
}
}
if (-not $MinGWBin) {
Write-ErrorExit "MinGW bin directory with gcc.exe not found. Tried: $($possibleBins -join ', ')"
}

$gccExe = Join-Path $MinGWBin "gcc.exe"
$gppExe = Join-Path $MinGWBin "g++.exe"
$arExe = Join-Path $MinGWBin "ar.exe"
if (!(Test-Path $gccExe)) { Write-ErrorExit "gcc.exe not found in $MinGWBin" }

Write-Host " MinGW bin directory: $MinGWBin" -ForegroundColor Gray

Force our MinGW to be first in PATH

$env:PATH = "$MinGWBin;$realBinDir;$env:PATH"

Override compiler detection environment variables

$env:CC = $gccExe
$env:CXX = $gppExe
$env:AR = $arExe
$env:CC_x86_64_pc_windows_gnu = $gccExe
${env:CC_x86_64-pc-windows-gnu} = $gccExe
$env:CXX_x86_64_pc_windows_gnu = $gppExe
${env:CXX_x86_64-pc-windows-gnu} = $gppExe
$env:AR_x86_64_pc_windows_gnu = $arExe
${env:AR_x86_64-pc-windows-gnu} = $arExe
$env:CARGO_TARGET_X86_64_PC_WINDOWS_GNU_LINKER = $gccExe
$env:CARGO_TARGET_X86_64_PC_WINDOWS_GNU_AR = $arExe

Write-Host " MinGW added to PATH (priority over Rust self-contained)." -ForegroundColor Gray
Write-Host " rustc version: $(& $rustcPath --version)" -ForegroundColor Gray
Write-Success "Environment ready."

--- 2. Locate Cargo cache ---

Write-Step "[2/7] Locating Cargo cache..."
$registryDir = Join-Path $RustDir "registry"
if (Test-Path $registryDir) {
Write-Success "Cache found at: $registryDir"
} else {
Write-Warning "No local Cargo cache found. Vendor may be required."
}

--- 3. Check / rebuild vendor directory ---

Write-Step "[3/7] Checking vendor directory..."
$vendorDir = Join-Path $RustDir "vendor"
$vendorValid = $false
if ((Test-Path $vendorDir) -and ((Get-ChildItem $vendorDir -Directory -ErrorAction SilentlyContinue).Count -gt 5)) {
Write-Success "Vendor directory exists and appears complete."
$vendorValid = $true
} else {
Write-Warning "Vendor missing or incomplete. Rebuilding from local cache..."
Write-Step "[3a/7] Rebuilding vendor from cache (offline)..."

Push-Location (Join-Path $SourceDir "rust")
$projectCargoDir = ".cargo"
if (Test-Path $projectCargoDir) { Remove-Item $projectCargoDir -Recurse -Force }
New-Item -ItemType Directory -Path $projectCargoDir -Force | Out-Null
@"

[source.crates-io]
registry = "https://github.com/rust-lang/crates.io-index"
"@ | Out-File (Join-Path $projectCargoDir "config.toml") -Encoding utf8

if (Test-Path $vendorDir) { Remove-Item $vendorDir -Recurse -Force }

Write-Host "       Running cargo vendor --offline..." -ForegroundColor Gray
$success = Invoke-CargoCommand -Command $cargoPath -Arguments "vendor --offline --versioned-dirs `"$vendorDir`"" -WorkingDirectory (Get-Location).Path -LogPrefix "cargo_vendor"
if (-not $success) { Write-ErrorExit "Vendor generation failed." }
Write-Success "Vendor rebuilt successfully."
Pop-Location
$vendorValid = $true

}

--- 4. Configure Cargo for offline build ---

Write-Step "[4/7] Configuring Cargo for offline build..."
Push-Location (Join-Path $SourceDir "rust")
$projectCargoDir = ".cargo"
if (-not (Test-Path $projectCargoDir)) { New-Item -ItemType Directory -Path $projectCargoDir -Force | Out-Null }
$configPath = Join-Path $projectCargoDir "config.toml"

$vendorPathFormatted = $vendorDir -replace '\', '/'
$gccPathFormatted = $gccExe -replace '\', '/'
$arPathFormatted = $arExe -replace '\', '/'

@"

[source.crates-io]
replace-with = "vendored-sources"

[source.vendored-sources]
directory = "$vendorPathFormatted"

[target.x86_64-pc-windows-gnu]
linker = "$gccPathFormatted"
ar = "$arPathFormatted"

[net]
offline = true
"@ | Out-File -FilePath $configPath -Encoding utf8

Write-Success "Configured vendored sources and forced MinGW tools."

--- 5. Compile Claw Code (offline) ---

Write-Step "[5/7] Compiling Claw Code (release) ..."
$ClawExe = Join-Path (Get-Location) "target\release\claw.exe"
if (Test-Path $ClawExe) {
$choice = Read-Host " Binary exists. Recompile? (y/N)"
if ($choice -ne 'y' -and $choice -ne 'Y') {
Write-Success "Skipping compilation."
$skipCompile = $true
} else {
$skipCompile = $false
Remove-Item "target" -Recurse -Force -ErrorAction SilentlyContinue
}
} else {
$skipCompile = $false
}

if (-not $skipCompile) {
Write-Host " Running cargo build --release --workspace --offline..." -ForegroundColor Gray
$success = Invoke-CargoCommand -Command $cargoPath -Arguments "build --release --workspace --offline" -WorkingDirectory (Get-Location).Path -LogPrefix "cargo_build"
if (-not $success) { Write-ErrorExit "Compilation failed." }
Write-Success "Compilation successful."
} else {
Write-Host " Using existing binary." -ForegroundColor Gray
}
$fileInfo = Get-Item $ClawExe
Write-Host " Binary: $ClawExe ($([math]::Round($fileInfo.Length / 1MB, 2)) MB)" -ForegroundColor Gray

--- 6. Configure .env for llama-server ---

Write-Step "[6/7] Configuring .env file..."
$EnvFile = Join-Path (Split-Path $ClawExe -Parent) ".env"
if (Test-Path $EnvFile) {
$choice = Read-Host " .env exists. Overwrite? (y/N)"
if ($choice -ne 'y' -and $choice -ne 'Y') {
Write-Success "Keeping existing .env."
$skipEnv = $true
} else {
$skipEnv = $false
}
} else {
$skipEnv = $false
}

if (-not $skipEnv) {
@"

Claw Code Configuration for Local llama-server

Generated by setup_clawcode_offline.ps1 v17.6

ANTHROPIC_BASE_URL=http://${LlamaserverHost}:${LlamaserverPort}
ANTHROPIC_API_KEY=local-llama-server-no-key-required
ANTHROPIC_MODEL=${LlamaserverModel}
CLAW_TOOL_EXECUTION=true
RUST_LOG=info
"@ | Out-File -FilePath $EnvFile -Encoding utf8
Write-Success ".env created with model: ${LlamaserverModel}"
}

--- 7. Final summary ---

Write-Step "[7/7] Finalizing..."
Pop-Location

Write-Host ""
Write-Host "========================================" -ForegroundColor Cyan
Write-Host "Offline setup completed successfully!" -ForegroundColor Green
Write-Host "========================================" -ForegroundColor Cyan
Write-Host "Binary: $ClawExe"
Write-Host "Config: $EnvFile"
Write-Host ""
Write-Host "To test Claw Code with your llama-server:"
Write-Host " 1. Ensure llama-server is running (as per your command):"
Write-Host " .\llama-server.exe -t 24 -c 1010000 --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 262144 --mlock --no-mmap -ngl 0 -m Qwen3-Coder-Next\80B\Qwen3-Coder-Next-Q4_K_M-00001-of-00004.gguf --jinja --offline -fa on --temp 0.3 --top-p 0.95 --top_k 40 --min-p 0 -n 32768 --no-context-shift --host 0.0.0.0 --port 8080"
Write-Host " 2. Navigate to the binary directory:"
Write-Host " cd $((Split-Path $ClawExe))"
Write-Host " 3. Run Claw Code doctor:"
Write-Host " .\claw.exe doctor"
Write-Host " 4. Run a test prompt:"
Write-Host " .\claw.exe prompt 'Hello, who are you?'"
Write-Host ""

[roflsunriz]

Don't configure this environment variables:
ANTHROPIC_BASE_URL=http://localhost:8080/
ANTHROPIC_API_KEY=local-llama-server-no-key-required

llama-server.exe is Open-AI Compatible API, so configure this:
OPENAI_BASE_URL=http://localhost:8080/v1
OPENAI_API_KEY=local-key

[roflsunriz]

I created guide so please follow this (if you don't know how to use claw code) :
https://roflsunriz.github.io/how-to-use-claw-code/

[TangLaoya]

Thank you very much!

After set the environment variables by your suggestion and run again, the claw still didn't work:

PS C:\AI_research\Quad_bipartite_label_AI> C:\claude_code\claw\clawcode\claw-code\rust\target\release\claw --model openapi/Qwen3.6-35B-A3B-Instruct
�[38;5;196m██████╗██╗ █████╗ ██╗ ██╗
██╔════╝██║ ██╔══██╗██║ ██║
██║ ██║ ███████║██║ █╗ ██║
██║ ██║ ██╔══██║██║███╗██║
╚██████╗███████╗██║ ██║╚███╔███╔╝
╚═════╝╚══════╝╚═╝ ╚═╝ ╚══╝╚══╝�[0m �[38;5;208mCode�[0m 🦞

�[2mModel�[0m openapi/Qwen3.6-35B-A3B-Instruct
�[2mPermissions�[0m danger-full-access
�[2mBranch�[0m unknown
�[2mWorkspace�[0m clean
�[2mDirectory�[0m C:\AI_research\Quad_bipartite_label_AI
�[2mSession�[0m session-1777507317706-0
�[2mAuto-save�[0m \?\C:\AI_research\Quad_bipartite_label_AI.claw\sessions\ff54a51b09285dd9\session-1777507317706-0.jsonl

Type �[1m/help�[0m for commands · �[1m/status�[0m for live context · �[2m/resume latest�[0m jumps back to the newest session · �[1m/diff�[0m then �[1m/commit�[0m to ship · �[2mTab�[0m for workflow completions · �[2mShift+Enter�[0m for newline
Connected: openapi/Qwen3.6-35B-A3B-Instruct via openai

请严格按照CLAUDE.md自动执行所有任务，执行完一条自动输出下一条，直到全部完成。
我来开始自动执行任务。首先读取任务清单和相关文件。<tool_call>
<function=read_file>
<parameter=path>
.claw/memory/tasks.md

.claw/memory/skills.md .claw/memory/modifycode.md ✔ ✨ Done

请继续执行tasks.md中定义的任务。
让我先读取任务清单和相关文件来了解当前状态。<tool_call>
<function=read_file>
<parameter=path>
.claw/memory/tasks.md

.claw/memory/skills.md .claw/memory/modifycode.md ✔ ✨ Done

It can be found that the claw only read the tasks.md, skills.md and modifycode.md as required in CLAUDE.md (the real work is defined in tasks.md), but didn't do any real operation to the workspace.

What else should I set/configure furthermore?

Thanks,
Tang Laoya

[roflsunriz]

I suggest you to try it with real frontier-capability models.
From my experience, I tried deepseek/deepdeek-v4-pro via OpenRouter but the model keeps getting fail to use tool calls (which is git commit and git push, considered as most easiest task), but then I have switched to openai/gpt-5.5, then all tool call success.
So I don't know what root cause was wrong but there are models that very good at terminal-based workflow and and others are not.

[TangLaoya]

Thank you very much for your kindly reply. I am required to configure the environment in offline machine. So I can't use the best models such as gpt-5.5, I can only use the local model by llama-server.

Could anyone suggest some better open source models that are good at terminal-based workflow?

Thanks

[roflsunriz]

At least based on what I've tried so far, all local models hallucinate and are not usable in a terminal-based workflow.
I personally think that unless it is a model that runs on workstation‑level GPUs, preventing such situations is difficult.

Tested models:
DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M
qwen2.5-coder-14b-instruct-q4_k_m
Qwen-3.5-27B-Derestricted.Q2_K

[roflsunriz]

gpt-oss-20b-Q8_0.gguf is best for my environment

[TangLaoya]

gpt-oss-20b-Q8_0.gguf is best for my environment

That's great! Thanks for your recommend, I will test it latter.

[TangLaoya]

I have configured the
GPTOSS-120B-Uncensored-HauhauCS-Aggressive-MXFP4.gguf
by
.\llama-server.exe -t 24 --mlock --no-mmap -ngl 0 -m .\GPTOSS-120B-Uncensored-HauhauCS-Aggressive-MXFP4.gguf --alias 'GPTOSS-120B' --chat-template --skip-chat-parsing --offline -fa on --temp 0.1 --top-p 0.95 --top-k 20 --min-p 0.01 --host 0.0.0.0 --port 8080 --ubatch-size 2048 --batch-size 2048 --cache-prompt

and then run the claw, the result is as follows, but the output is TOTALLY not what I wanted. So it seems that the LLM really doesn't 'understand' what the claw 'say'.

PS C:\AI_research\Quad_bipartite_label_AI> claw --model openapi/GPSOSS-120B
�[38;5;196m██████╗██╗ █████╗ ██╗ ██╗
██╔════╝██║ ██╔══██╗██║ ██║
██║ ██║ ███████║██║ █╗ ██║
██║ ██║ ██╔══██║██║███╗██║
╚██████╗███████╗██║ ██║╚███╔███╔╝
╚═════╝╚══════╝╚═╝ ╚═╝ ╚══╝╚══╝�[0m �[38;5;208mCode�[0m 🦞

�[2mModel�[0m openapi/GPSOSS-120B
�[2mPermissions�[0m danger-full-access
�[2mBranch�[0m unknown
�[2mWorkspace�[0m clean
�[2mDirectory�[0m C:\AI_research\Quad_bipartite_label_AI
�[2mSession�[0m session-1777715231162-0
�[2mAuto-save�[0m \?\C:\AI_research\Quad_bipartite_label_AI.claw\sessions\ff54a51b09285dd9\session-1777715231162-0.jsonl

Type �[1m/help�[0m for commands · �[1m/status�[0m for live context · �[2m/resume latest�[0m jumps back to the newest session · �[1m/diff�[0m then �[1m/commit�[0m to ship · �[2mTab�[0m for workflow completions · �[2mShift+Enter�[0m for newline
Connected: openapi/GPSOSS-120B via openai

请严格按照CLAUDE.md自动执行所有任务，执行完一条自动输出下一条，直到全部完成。
flag is set, the script will skip the chat parsing step and directly use the existingchat.jsonfile.╭─ bash python script.py --skip-chat-parsing ╰─2. Running the Script with the--skip-embeddingFlagIf you want to skip the embedding step and use the existingembeddings.npyfile, you can use the--skip-embeddingflag.╭─ bash python script.py --skip-embedding ╰─3. Running the Script with Both FlagsYou can also combine both flags if you want to skip both the chat parsing and embedding steps.╭─ bash python script.py --skip-chat-parsing --skip-embedding ╰─4. Running the Script with a Custom Model NameIf you want to use a different model for embedding, you can specify the model name using the--model_nameargument.╭─ bash python script.py --model_name "sentence-transformers/all-MiniLM-L6-v2" ╰─5. Running the Script with a Custom Output DirectoryIf you want to specify a different output directory for the results, you can use the--output_dirargument.╭─ bash python script.py --output_dir "custom_output" ╰─6. Running the Script with a Custom Number of ClustersIf you want to specify a different number of clusters for the KMeans clustering, you can use the--num_clustersargument.╭─ bash python script.py --num_clusters 5 ╰─7. Running the Script with a Custom Number of TopicsIf you want to specify a different number of topics for the BERTopic model, you can use the--num_topicsargument.╭─ bash python script.py --num_topics 15 ╰─8. Running the Script with a Custom Number of Top WordsIf you want to specify a different number of top words to extract from each topic, you can use the--top_n_wordsargument.╭─ bash python script.py --top_n_words 20 ╰─9. Running the Script with a Custom Number of Top SentencesIf you want to specify a different number of top sentences to extract from each topic, you can use the--top_n_sentencesargument.╭─ bash python script.py --top_n_sentences 5 ╰─10. Running the Script with a Custom Number of Top ImagesIf you want to specify a different number of top images to extract from each topic, you can use the--top_n_imagesargument.╭─ bash python script.py --top_n_images 5 ╰─11. Running the Script with a Custom Number of Top FilesIf you want to specify a different number of top files to extract from each topic, you can use the--top_n_filesargument.╭─ bash python script.py --top_n_files 5 ╰─12. Running the Script with a Custom Number of Top Code SnippetsIf you want to specify a different number of top code snippets to extract from each topic, you can use the--top_n_codeargument.╭─ bash python script.py --top_n_code 5 ╰─13. Running the Script with a Custom Number of Top LinksIf you want to specify a different number of top links to extract from each topic, you can use the--top_n_linksargument.╭─ bash python script.py --top_n_links 5 ╰─14. Running the Script with a Custom Number of Top ImagesIf you want to specify a different number of top images to extract from each topic, you can use the--top_n_images` argument.╭─ bash
python script.py --top_n_images 5
╰─

Is there still anything I have missed?

Thanks,
Tang Laoya

[roflsunriz]

Can you try it with opencode?
You can install it via

npm install -g opencode-ai@latest

or

bun install -g opencode-ai@latest

or

yarn or pnpm, whatever your preferred package managers.

then

Create opencode.json on $env:USERPROFILE\.config\opencode

as follows:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "local-ai": {
      "api": "http://127.0.0.1:8080",
      "npm": "@ai-sdk/openai-compatible",
      "models": {
		"GPTOSS-120B-Uncensored-HauhauCS-Aggressive-MXFP4" : { "name": "Local-GPT-OSS-120B"}
      }
    }
  }
}

follow the Guide.
Maybe it will work then claw-code-specific problem.
Otherwise, model-specific problem.

[TangLaoya]

Thanks for your suggestion. Do you mean that I can use the opencode to check and solve the claw-code specific problem?
I have tried the opencode several months ago, but I have failed to launch it on my offline Windows 10 machine, so I gave up.

[roflsunriz]

yeah, you don't need to use opencode, just use other cli agent and tell me what output looks like.
Also, llama-server has built-in WebUI and tell me what output is looks like also.

If you use other cli agent and models can use function call normally then claw-code problem.

[TangLaoya]

Oh, it seems that there are some problems in the following command?
.\llama-server.exe -t 24 --mlock --no-mmap -ngl 0 -m .\GPTOSS-120B-Uncensored-HauhauCS-Aggressive-MXFP4.gguf --alias 'GPTOSS-120B' --chat-template --skip-chat-parsing --offline -fa on --temp 0.1 --top-p 0.95 --top-k 20 --min-p 0.01 --host 0.0.0.0 --port 8080 --ubatch-size 2048 --batch-size 2048 --cache-prompt

when I use the WebUI to ask question, it aslo display similar message:

` flag is set, the script will skip the chat parsing step and only generate the HTML file.

If you want to use a different template file, you can specify the path to the template file using the --template flag. For example:

bash

python script.py --template path/to/template.html
If you want to use a different output file name, you can specify the name using the --output flag. For example:

bash

python script.py --output my_output.html
If you want to use a different input file name, you can specify the name using the --input flag. For example:

bash

python script.py --input my_input.txt
Sure! Here's a revised version of the script that includes the changes you requested:

python
...

How to modify the command?

Thanks,
Tang Laoya

[roflsunriz]

By the way Opencode is just working fine with offline