RF3 error ConvertTritonGPUToLLVM

Hi There
First of all thanks for your great work! I am getting an error when running structure prediction with rf3. I am using this command:
rf3 fold inputs=test_input/test_struct.pdb ckpt_path=./checkpoints/rf3_foundry_01_24_latest_remapped.ckpt out_dir=test_pdb

And this is my error:

Unsupported conversion from bf16 to f16 LLVM ERROR: Unsupported rounding mode for conversion. #blocked = #ttg.blocked<{sizePerThread = [4, 4], threadsPerWarp = [4, 8], warpsPerCTA = [4, 1], order = [1, 0]}> #blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [8, 4], warpsPerCTA = [4, 1], order = [1, 0]}> #blocked2 = #ttg.blocked<{sizePerThread = [8, 1], threadsPerWarp = [8, 4], warpsPerCTA = [1, 4], order = [0, 1]}> #blocked3 = #ttg.blocked<{sizePerThread = [8, 1], threadsPerWarp = [4, 8], warpsPerCTA = [1, 4], order = [0, 1]}> #shared = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [1, 0]}> #shared1 = #ttg.swizzled_shared<{vec = 1, perPhase = 1, maxPhase = 1, order = [0, 1]}> #smem = #ttg.shared_memory module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, ttg.target = "cuda:70", "ttg.threads-per-warp" = 32 : i32} { tt.func public @fused_sigmoid_gated_dual_gemm_forward_kernel(%arg0: !tt.ptr<bf16> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<bf16> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<bf16> {tt.divisibility = 16 : i32}, %arg3: i32 {tt.divisibility = 16 : i32}, %arg4: i32 {tt.divisibility = 16 : i32}, %arg5: i32 {tt.divisibility = 16 : i32}, %arg6: !tt.ptr<bf16> {tt.divisibility = 16 : i32}) attributes {noinline = false} {
...
{-# external_resources: { mlir_reproducer: { pipeline: "builtin.module(triton-nvidia-mma-lowering, tritongpu-combine-tensor-select-and-if, tritongpu-allocate-warp-groups, convert-scf-to-cf, allocate-shared-memory, triton-tensor-memory-allocation, tritongpu-global-scratch-memory-allocation, convert-triton-gpu-to-llvm{compute-capability=70 ptx-version=84}, canonicalize{ max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}, cse, convert-nv-gpu-to-llvm, convert-warp-specialize-to-llvm, canonicalize{ max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}, cse, symbol-dce, enable-line-info)", disable_threading: false, verify_each: true } } #-} /work/ps40kyji-RFD3/foundry/.venv/lib/python3.12/site-packages/cuequivariance_ops/triton/gated_gemm_triton.py:19:0: error: Failures have been detected while processing an MLIR pass pipeline /work/ps40kyji-RFD3/foundry/.venv/lib/python3.12/site-packages/cuequivariance_ops/triton/gated_gemm_triton.py:19:0: note: Pipeline failed while executing [ConvertTritonGPUToLLVMon 'builtin.module' operation]: reproducer generated atstd::errs, please share the reproducer above with Triton project.Traceback (most recent call last):
What could cause this issue?
Thanks very much in advance!

Reproduce the failure with the documented rf3 fold command and inspect cuequivariance_ops/triton/gated_gemm_triton.py:19 along with the generated MLIR reproducer. Check the Triton, CUDA, and checkpoint/runtime compatibility involved in ConvertTritonGPUToLLVM. Done means identifying the cause of the bf16-to-f16 conversion failure and documenting a verified resolution or compatibility requirement.