add nosupport K100 FA code

ktransformers/local_chat.py

@@ -171,7 +171,7 @@ def local_chat(
             "please change max_seq_len in  ~/.ktransformers/config.yaml"
         
         #if system != "Windows" and (config.architectures[0] == "DeepseekV2ForCausalLM" or config.architectures[0] == "DeepseekV3ForCausalLM") and flashinfer_enabled and get_compute_capability() >= 8:
-        if system != "Windows" and (config.architectures[0] == "DeepseekV2ForCausalLM" or "DeepseekV3ForCausalLM") and flashinfer_enabled and (get_compute_capability() >= 8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name())):
+        if system != "Windows" and (config.architectures[0] == "DeepseekV2ForCausalLM" or "DeepseekV3ForCausalLM") and flashinfer_enabled and (get_compute_capability() >= 8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name())) or ("K100" in get_device_name()):
             generated = prefill_and_generate(
                 model, tokenizer, input_tensor.cuda(), max_new_tokens, use_cuda_graph, mode = mode, force_think = force_think, chunk_prefill_size = chunk_prefill_size,
                 use_flashinfer_mla = True, num_heads = config.num_attention_heads, head_dim_ckv = config.kv_lora_rank, head_dim_kpe = config.qk_rope_head_dim, q_head_dim = config.qk_rope_head_dim + config.qk_nope_head_dim

ktransformers/operators/attention.py

@@ -592,8 +592,8 @@ class KDeepseekV2Attention(BaseInjectedModule, DeepseekV2Attention):
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
         #if os.name == 'nt' or get_compute_capability()<8:
             #print("for Windows or GPU before ampere, use forward_windows")
-        if os.name == 'nt' or get_compute_capability()<8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name()):
-            print("for Windows or GPU before ampere or Z100/Z100L, use forward_windows")
+        if os.name == 'nt' or get_compute_capability()<8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name()) or ("K100" in get_device_name()):
+            print("for Windows or GPU before ampere or Z100/Z100L or K100, use forward_windows")
             return self.forward_windows(
                 hidden_states,
                 attention_mask,

ktransformers/operators/models.py

@@ -652,8 +652,8 @@ class KDeepseekV2Model(BaseInjectedModule):
         else:
             #if os.name == 'nt' or get_compute_capability()<8:
              #   print("for Windows or GPU before ampere, use forward_windows")
-            if os.name == 'nt' or get_compute_capability()<8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name()):
-                print("for Windows or GPU before ampere or Z100/Z100L, use forward_windows")
+            if os.name == 'nt' or get_compute_capability()<8 or ("Z100" in get_device_name()) or ("Z100L" in get_device_name()) or ("K100" in get_device_name()):
+                print("for Windows or GPU before ampere or Z100/Z100L or K100, use forward_windows")
                 # only use mask in forward windows or can't flash attn
                 causal_mask = self._update_causal_mask(
                     attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions