








llama-guard-3-8b input schema is just a copy paste of every other model page but it doesn't actually match what is live. For example:system and tool are not valid message rolespyannote on cloudflare?
guidance to something like 1, possibly 2. With higher values, problems with saturation and contrast quickly become apparent. The img2img and inpainting models had similar issues last time I checked. (See cloudflare-ai for example.) And SDXL-lightning has a somewhat similar issue. (Cartoonish speckled result unless you set num_steps to e.g. 2 and guidance to a low value.)ai/run endpoint truncated when @cf/deepseek-ai/deepseek-r1-distill-qwen-32b is clearly midsentence. The request is always 200 too. Is this a known issue of some kind? This has been happening for the past 2 days now.workerd/server/workerd-api.c++:759: error: wrapped binding module can't be resolved (internal modules only); moduleName = miniflare-internal:wrapped:__WRANGLER_EXTERNAL_AI_WORKER
workerd/jsg/util.c++:331: error: e = workerd/server/workerd-api.c++:789: failed: expected !value.IsEmpty(); global did not produce v8::Value