Bug: Clip-L does absolutely nothing with Flux models.

[stduhpf]

I was testing out alternative clip models for Flux (https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14), and noticed the output were oddly similar to the ones I was getting with the original clip-l.
Turns out the result are actually 100% identical (when using the same seed of course), not even a single bit of difference. Using the same models with ComfyUI does lead to noticably different outputs (with fixed seed).

	clip-L	ViT-L-14
Dev q3_k (sd.cpp)
Schnell q3_k (sd.cpp)
Schnell fp8 (Comfyui)
Schnell q3_k (Comfyui)

As a sidenote, this could explain why I was feeling like the generations I get with sdcpp look less detailed and more artifacty than when I'm using ComfyUI at the same quants.

[Green-Sky]

You are right, it is commented out and just set to 0.

stable-diffusion.cpp/conditioner.hpp

Lines 1107 to 1121 in 14206fd

	// auto it = std::find(chunk_tokens.begin(), chunk_tokens.end(), clip_l_tokenizer.EOS_TOKEN_ID);
	// max_token_idx = std::min<size_t>(std::distance(chunk_tokens.begin(), it), chunk_tokens.size() - 1);
	// clip_l->compute(n_threads,
	// input_ids,
	// 0,
	// NULL,
	// max_token_idx,
	// true,
	// &pooled,
	// work_ctx);
	// clip_l.transformer.text_model.text_projection no in file, ignore
	// TODO: use torch.eye(embed_dim) as default clip_l.transformer.text_model.text_projection
	pooled = ggml_new_tensor_1d(work_ctx, GGML_TYPE_F32, 768);
	ggml_set_f32(pooled, 0.f);