#3439 Strata on RX 6900 XT (gfx1030) without compiling the HIP engine myself?

Description

I'm trying to get Strata running on Windows 11 with my RX 6900 XT. The quickstart detects the GPU successfully, but setup stops because no prebuilt AMD engine is available.

My system

GPU: AMD Radeon RX 6900 XT, 16 GB VRAM (gfx1030)
CPU: AMD Ryzen 5 5600X, 6 cores
RAM: 128 GB DDR4
OS: Windows 11
Strata version: v0.1.46.1
Selected model: Qwen3.8-Flash-Next
Quantization: IQ3_S
Context length: 65,536 tokens
KV cache: 8-bit

I selected the original model with IQ3_S, 64K context and 8-bit KV cache. Python dependencies installed successfully. The setup then stopped at Step 4.

Full relevant console output:

=== Step 1: checking your PC ===
  Your AMD GPUs:
    GPU 0: AMD Radeon RX 6900 XT, 16 GB VRAM - can be used
  [ok] GPU: AMD Radeon RX 6900 XT, 16.0 GB VRAM, gfx1030 (AMD: docs/AMD_HIP.md)
  [ok] RAM: 128 GB
  [ok] CPU: AMD Ryzen 5 5600X 6-Core Processor (AVX2)

=== Step 2: your choices ===
  1) Qwen3.8-Flash-Next   Qwen; GSQ-RCO quants by ISTA-DASLab - the original model
  2) Swift 1.5            UkisAI's fine-tune of Qwen3.8-Flash-Next - thinks much shorter (-63% thinking tokens, 1.8x sooner answers by its authors' numbers)
  3) Qwen3.8-Flash-Next Coder ISTA-DASLab's coding version - half the experts (code, tools, images kept): needs ~32 GB of RAM, faster; weaker outside coding
  4) Qwen3.8-Flash-Next (Unsloth) Unsloth's 4- to 8-bit quantizations - UD-IQ4_XS, UD-Q4_K_XL, UD-Q5_K_XL, UD-Q6_K_XL or Q8_0 (94 to 188 GB downloads): every expert in RAM from ~80 GB (UD-IQ4_XS) to ~136 GB (Q8_0) of RAM, else part of them read from the SSD   [UD-IQ4_XS is the 4-bit choice on AMD and Intel cards; NVFP4 runs on NVIDIA only]
  5) Qwen3.8-Flash-Next (Gyro) Gyro quantizations by agentionai - Gyro-S (27.6 GB of memory) and Gyro-M (35 GB): the smallest sizes, for PCs with little memory; 58 and 92 GB downloads (the files include the n-gram table)   [for PCs with little memory: VRAM + RAM together 27.6 GB (Gyro-S) or 35 GB (Gyro-M); AMD is experimental, not for Intel Arc]
  6) Qwen3.6-35B-A3B      Qwen; GGUFs with the MTP layer by Unsloth - 35B mixture of experts with 3B active (256 experts, top-8): a smaller cousin of Qwen3.8-Flash-Next with the same tokenizer   [new]
Which model? [1]:
  [ok] model: Qwen3.8-Flash-Next

  1) Q2_0     2-bit, the fastest; download 66 GB, uses ~34 GB of RAM
  2) IQ2_XS   2-bit i-quant, a little better quality, close in speed; download 68 GB, uses ~36 GB of RAM
  3) IQ3_XXS  3-bit i-quant, better quality, slower (more CPU work per token); download 76 GB, uses ~43 GB of RAM
  4) IQ3_S    (recommended) 3.5-bit i-quant, the best quality (matches the full model), the slowest; needs a 64 GB PC with little else running; download 84 GB, uses ~50 GB of RAM
Which size? [4]: 4

  Context length = how much text the model can see at once (your chat, files, tool output).
  Longer needs more VRAM for it, so fewer experts fit on the GPU:
  1) 8K tokens
  2) 32K tokens
  3) 64K tokens   (recommended for your GPU)
  4) 128K tokens
  5) 200K tokens
  6) 256K tokens
  7) 384K tokens   (experimental: setup adds rope scaling)
  8) 512K tokens   (experimental: setup adds rope scaling)
Context? [3]: 3
  [ok] context: 65536 tokens

  KV cache precision (the model's memory of the conversation):
  1) 8-bit   (recommended: what every published number was measured with)
  2) 4-bit   half the memory (about 4% faster at 128K), but measurably less precise on long
             documents; long-context lookups (needle tests) still pass
KV cache? [1]: 1
  [ok] KV cache: 8-bit
  [ok] images: on (encoder on the CPU)

  EXPERIMENTAL - speed projection: a small control vector applied while the model runs (layers 4-44).
  It changes how the model answers: its package describes it as a refusal-direction projection (the
  model declines far fewer requests). Off unless you choose it; when on, the web app can switch it off
  per chat. Details: data/experimental-speed-projection/README.md
Turn on the experimental speed projection? [n]:
  [ok] experimental speed projection: off

=== Step 3: Python packages ===
  Installing numpy, jinja2, regex, pyyaml, tqdm, requests, pillow, psutil ...
  > S:\AI\Strata\.venv\Scripts\python.exe -m pip install --quiet --disable-pip-version-check numpy==2.5.3; python_version >= "3.12" numpy==2.4.6; python_version == "3.11" numpy==2.2.6; python_version < "3.11" jinja2==3.1.6 regex==2026.9.10 pyyaml==6.0.3 tqdm==4.70.1 requests==2.34.2 pillow==12.3.0 psutil==7.2.2 markupsafe==3.0.3 certifi==2026.7.22 charset-normalizer==3.5.1 idna==3.20 urllib3==2.8.0 colorama==0.4.6; sys_platform == "win32"
  [ok] numpy, jinja2, regex, pyyaml, tqdm, requests, pillow, psutil installed

=== Step 4: the Strata engine ===
  llama.cpp source:     8.4 MB
  [ok] llama.cpp source downloaded
  [ok] llama.cpp 3cf0325 (gguf-py, ggml, mtmd)
  No ready-made AMD engine for v0.1.46.1 (HTTP Error 404: Not Found): the latest release instead
  [!]  no ready-made AMD engine at https://github.com/Niko1221/Strata/releases/latest/download/ (HTTP Error 404: Not Found)

  [X]  no ready-made AMD engine for this Strata version
       compiling it on Windows: tools\hip\build_windows.bat makes strata-windows-x64-hip.zip, then run START-HERE.bat --backend hip --prebuilt <its dist folder> (docs/AMD_HIP.md)

Setup stopped. Fix the item above and run it again - everything already done is kept and skipped.
Drücken Sie eine beliebige Taste . . .


I found RCOm-windows-gfx1030, which appears to provide a Windows ROCm/PyTorch setup for gfx1030.

Has anyone successfully used this setup, or any of its components, to run Strata on an RX 6900 XT?

I'm looking for a working path using existing binaries or packages, not a theoretical compatibility discussion. If RCOm cannot help because Strata needs its own GGML/llama.cpp HIP backend, is there another known working Strata release, binary, or configuration for gfx1030?

Can I use an existing compatible engine or package instead of compiling the HIP engine and its dependencies myself?

Thanks!