Qualcomm AI Engine Direct - GLM1.5B (#15691)

### Summary

GLM Enablement
`python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s
$DEVICE -m SM8750 --temperature 0 --model_mode kv --max_seq_len 128
--decoder_model glm-1_5b --prompt "Could you tell me about Facebook?"`

### Test plan
`python backends/qualcomm/tests/test_qnn_delegate.py -k
TestExampleLLMScript.test_static_glm1_5b --model SM8750 --build_folder
build-android/ --executorch_root . -s $DEVICE --artifact ./glm1_5b`

Author: winskuo-quic

backends/qualcomm/tests/test_qnn_delegate.py

@@ -5862,6 +5862,9 @@ def setUp(self):
             "gemma3-1b": TestExampleLLMScript.LlmSpecs(
                 SM8650=70, SM8750=100, ppl=23, pte_size=1_200_000_000
             ),  # 1.2 GB
+            "glm-1_5b": TestExampleLLMScript.LlmSpecs(
+                SM8650=42, SM8750=52, ppl=21, pte_size=1_100_000_000
+            ),  # 1.1 GB
             "phi_4_mini": TestExampleLLMScript.LlmSpecs(
                 SM8650=14, SM8750=19, ppl=12, pte_size=4_000_000_000
             ),  # 4GB

examples/models/glm/__init__.py

@@ -0,0 +1,16 @@
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from executorch.examples.models.glm.convert_weights import convert_weights
+from executorch.examples.models.llama.model import Llama2Model
+
+
+class GLMModel(Llama2Model):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+
+__all__ = [
+    "GLMModel",
+    "convert_weights",
+]

examples/models/glm/config/1_5b_config.json

@@ -0,0 +1,17 @@
+{
+  "dim": 2048,
+  "ffn_dim_multiplier": 1,
+  "hidden_dim": 6144,
+  "n_heads": 16,
+  "head_dim": 128,
+  "n_kv_heads": 4,
+  "n_layers": 28,
+  "norm_eps": 1e-05,
+  "rope_theta": 10000.0,
+  "use_scaled_rope": false,
+  "vocab_size": 59264,
+  "use_hf_rope": true,
+  "attention_qkv_bias": false,
+  "use_qk_norm": false,
+  "model_architecture" : "GlmForCausalLM"
+}

examples/models/glm/convert_weights.py

@@ -0,0 +1,79 @@
+import argparse
+import os
+from typing import Dict
+
+import torch
+from safetensors.torch import load_file
+from torchtune.models.convert_weights import get_mapped_key
+
+# Standard _FROM_META weight mapping of Meta weights to TorchTune + additional bias weight mappings.
+_GLM_FROM_META = {
+    "tok_embeddings.weight": "model.embed_tokens.weight",
+    "norm.weight": "model.norm.weight",
+    "output.weight": "lm_head.weight",
+    "layers.{}.attention.wk.weight": "model.layers.{}.self_attn.k_proj.weight",
+    "layers.{}.attention.wq.weight": "model.layers.{}.self_attn.q_proj.weight",
+    "layers.{}.attention.wv.weight": "model.layers.{}.self_attn.v_proj.weight",
+    "layers.{}.attention.wo.weight": "model.layers.{}.self_attn.o_proj.weight",
+    "layers.{}.attention_norm.weight": "model.layers.{}.input_layernorm.weight",
+    "layers.{}.ffn_norm.weight": "model.layers.{}.post_attention_layernorm.weight",
+    "layers.{}.feed_forward.gate_up_proj.weight": "model.layers.{}.mlp.gate_up_proj.weight",
+    "layers.{}.feed_forward.down_proj.weight": "model.layers.{}.mlp.down_proj.weight",
+}
+
+
+def glm_tune_to_meta(state_dict: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]:
+    """
+    Convert a state dict from torchtune's format to Meta's format. This function
+    doesn't handle any sharding or splitting of state dicts. It follows the
+    state_dict IN -> state_dict OUT pattern.
+
+    Args:
+        state_dict (Dict[str, torch.Tensor]): State dict in torchtune's format.
+
+    Returns:
+        Dict[str, torch.Tensor]: State dict in Meta's format.
+    """
+    converted_state_dict = {}
+    inverted_mapping_dict = {v: k for k, v in _GLM_FROM_META.items()}
+
+    for key, value in state_dict.items():
+        new_key = get_mapped_key(key, inverted_mapping_dict)
+        converted_state_dict[new_key] = value
+
+    if "lm_head.weight" not in state_dict:
+        converted_state_dict["output.weight"] = converted_state_dict[
+            "tok_embeddings.weight"
+        ]
+
+    return converted_state_dict
+
+
+def convert_weights(input_dir: str, output_file: str) -> None:
+    pt_path = os.path.join(input_dir, "model.safetensors")
+    print("Loading checkpoint from file...")
+    sd = load_file(pt_path)
+
+    print("Converting checkpoint...")
+    sd = glm_tune_to_meta(sd)
+
+    print("Saving checkpoint...")
+    torch.save(sd, output_file)
+    print("Done.")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Convert GLM weights to Meta format.")
+    parser.add_argument(
+        "input_dir",
+        type=str,
+        help="Path to directory containing checkpoint files",
+    )
+    parser.add_argument("output", type=str, help="Path to the output checkpoint")
+
+    args = parser.parse_args()
+    convert_weights(args.input_dir, args.output)
+
+
+if __name__ == "__main__":
+    main()

examples/models/llama/model_args.py

@@ -131,6 +131,9 @@ class ModelArgs:
     attention_kwargs: Dict[str, Any] = dataclasses.field(default_factory=dict)
     # Hybrid models can have layer types different from attention
     layer_types: Optional[list] = None
+    model_architecture: Optional[str] = (
+        None  # Architecture of model. For HF models, please refer to the HF model.config.architectures. This is used in QNN backend only for now.
+    )
 
     def __post_init__(self):
         if self.n_kv_heads is None:

examples/qualcomm/oss_scripts/llama/README.md

@@ -9,6 +9,7 @@ This file provides you the instructions to run LLM Decoder model with different
  1. Codegen2 1B
  1. Gemma 2B
  1. Gemma3 1B
+ 1. GLM 1.5B
  1. Granite3.3 2B
  1. Phi4-mini-instruct
  1. QWEN2.5 0.5B / 1.5B
@@ -65,7 +66,10 @@ Follow the [instructions](https://www.llama.com/) to download models.
 At the end of this step, users should have the following files ready: `consolidated.00.pth`, `params.json`, and `tokenizer.model`.
 
 
-### Step3: Run default examples using hybrid mode for smaller models and kv mode for larger models.
+### Step3: Run default examples.
+#### Note:
+All example scripts below use hybrid mode, which is optimized for on-device performance. However, compiling a model in hybrid mode can consume a significant amount of memory on the host machine—sometimes up to ~100 GB. If your host machine has limited memory, it is highly recommended to switch from `--model_mode hybrid` to `--model_mode kv` and remove the `--prefill_ar_len` flag.
+
 #### LLAMA2
 ```bash
 python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint stories110M.pt --params params.json --tokenizer_model tokenizer.model --tokenizer_bin tokenizer.bin --decoder_model stories110m --model_mode hybrid --prefill_ar_len 32 --max_seq_len 128 --prompt "Once upon a time"
@@ -80,7 +84,7 @@ python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL
 #### LLAMA3.2 3B Instruct
 Default example using kv mode.
 ```bash
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --checkpoint consolidated.00.pth --params params.json --tokenizer_model tokenizer.model --decoder_model llama3_2-3b_instruct --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
 #### Codegen2
@@ -102,6 +106,12 @@ Default example using hybrid mode
 python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model gemma3-1b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
+#### GLM 1.5B
+Default example using hybrid mode
+```bash
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --max_seq_len 1024 --prefill_ar_len 128 --decoder_model glm-1_5b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+```
+
 #### Granite3.3 2B
 Default example using hybrid mode
 ```bash
@@ -111,7 +121,7 @@ python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL
 #### Phi4-mini-instruct
 Default example using kv mode.
 ```bash
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --decoder_model phi_4_mini --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --decoder_model phi_4_mini --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
 #### QWEN2.5 0.5B
@@ -123,7 +133,7 @@ python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL
 #### QWEN2.5 1.5B
 Default example using kv mode
 ```bash
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode kv --max_seq_len 1024 --decoder_model qwen2_5-1_5b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
 #### QWEN3 0.6B
@@ -135,7 +145,7 @@ python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL
 #### QWEN3 1.7B
 Default example using hybrid mode
 ```bash
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode kv --max_seq_len 1024 --decoder_model qwen3-1_7b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --temperature 0 --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --decoder_model qwen3-1_7b --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
 #### SmolLM2
@@ -147,7 +157,7 @@ python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL
 #### SmolLM3
 Default example using kv mode.
 ```bash
-python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --decoder_model smollm3-3b --model_mode kv --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
+python examples/qualcomm/oss_scripts/llama/llama.py -b build-android -s ${SERIAL_NUM} -m ${SOC_MODEL} --decoder_model smollm3-3b --model_mode hybrid --prefill_ar_len 128 --max_seq_len 1024 --prompt "I would like to learn python, could you teach me with a simple example?" --tasks wikitext --limit 1
 ```
 
 ### KV Cache update mechanism

examples/qualcomm/oss_scripts/llama/__init__.py

@@ -16,6 +16,8 @@
 )
 from executorch.examples.models.gemma import convert_weights as convert_gemma_weights
 from executorch.examples.models.gemma3 import convert_weights as convert_gemma3_weights
+
+from executorch.examples.models.glm import convert_weights as convert_glm_weights
 from executorch.examples.models.granite import (
     convert_weights as convert_granite_weights,
 )
@@ -44,6 +46,7 @@
     CodegenQuantRecipe,
     Gemma3QuantRecipe,
     Gemma_2BQuantRecipe,
+    GLM_1_5B_InstructQuantRecipe,
     Granite_3_3_2B_InstructQuantRecipe,
     Llama3_1BQuantRecipe,
     Llama3_3BQuantRecipe,
@@ -293,6 +296,26 @@ class Gemma3(LLMModelConfig):
     quant_recipe = Gemma3QuantRecipe
 
 
+@register_llm_model("glm-1_5b")
+@dataclass(init=False, frozen=True)
+class GLM_1_5B(LLMModelConfig):
+    repo_id: str = "THUDM/glm-edge-1.5b-chat"
+    params_path: str = os.path.join(
+        BASE_DIR, "../../../models/glm/config/1_5b_config.json"
+    )
+    convert_weights = convert_glm_weights
+    transform_weight = True
+    instruct_model = True
+    num_sharding = 1
+    group_size = 32
+    masked_softmax = False
+    seq_mse_candidates = 0
+    r1 = False
+    r2 = False
+    r3 = False
+    quant_recipe = GLM_1_5B_InstructQuantRecipe
+
+
 @register_llm_model("granite_3_3-2b_instruct")
 @dataclass(init=False, frozen=True)
 class Granite_3_3_2b_Instruct(LLMModelConfig):

examples/qualcomm/oss_scripts/llama/decoder_constants.py

@@ -27,4 +27,5 @@
     "smollm2_135m": "smollm2_135m",
     "smollm3-3b": "smollm3",
     "codegen2_1b": "codegen",
+    "glm-1_5b": "glm",
 }

examples/qualcomm/oss_scripts/llama/llama.py

@@ -1309,15 +1309,23 @@ def export_llama(args) -> None:
             # For Gemma, use tokenizer.model as it doesn't provide pre_tokenizer in tokenizer.json.
             runtime_tokenizer_path = tokenizer_artifacts[-3]
         else:
+            if args.decoder_model == "glm-1_5b":
+                with open(tokenizer_config, "r+") as file:
+                    data = json.load(file)
+                    # Verified with HF flow and it uses <|user|> as eos condition
+                    data["bos_token"] = "<|user|>"
+                    data["eos_token"] = "<|user|>"
+                    file.seek(0)
+                    json.dump(data, file, indent=4)
+                    file.truncate()
             runtime_tokenizer_path = tokenizer_artifacts[-1]
+
         tokenizer = get_tokenizer(runtime_tokenizer_path, tokenizer_config)
 
     if args.decoder_model == "codegen2_1b":
         # Override the default BOS and EOS token IDs for codegen2_1b
         tokenizer.bos_id = 1
         tokenizer.eos_id = 2
-
-    # TODO: Remove this once error is resolved.
     elif args.decoder_model == "phi_4_mini":
         with open(runtime_tokenizer_path, "r+") as file:
             data = json.load(file)

examples/qualcomm/oss_scripts/llama/model/feed_forward.py

@@ -88,3 +88,54 @@ def forward(self, x):
         hidden_states = self.act(hidden_states)
         hidden_states = self.fc_out(hidden_states)
         return hidden_states
+
+
+@register_feed_forward("GlmForCausalLM")
+class GLMFeedForward(FeedForwardBase):
+    """FeedForward with gate_up_proj and down_proj"""
+
+    def __init__(self, args: ModelArgs):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+
+        assert args.hidden_dim is not None
+        self.dim = args.dim
+        self.hidden_dim = args.hidden_dim
+
+        self.gate_up_proj = torch.nn.Linear(args.dim, 2 * args.hidden_dim, bias=False)
+        self.down_proj = torch.nn.Linear(args.hidden_dim, args.dim, bias=False)
+        self.activation_fn = args.act_fn.get_function()
+
+    def prepare_feedfoward_conv(self):
+        self.gate_up_proj_conv = torch.nn.Conv2d(
+            self.dim, 2 * self.hidden_dim, 1, bias=False
+        )
+        self.down_proj_conv = torch.nn.Conv2d(self.hidden_dim, self.dim, 1, bias=False)
+
+        self.forward_no_conv = self.forward
+        self.forward = self.forward_feedfoward_conv
+
+        self.gate_up_proj_conv.weight.data.copy_(
+            self.gate_up_proj.weight[:, :, None, None]
+        )
+        self.down_proj_conv.weight.data.copy_(self.down_proj.weight[:, :, None, None])
+
+        del self.gate_up_proj
+        del self.down_proj
+
+    def forward_feedfoward_conv(self, x):
+        bsz, _, _ = x.size()
+        x = torch.reshape(x, (bsz, -1, 1, self.dim))
+        x = x.transpose(1, 3)  # Transpose right before and after Conv
+        up_states = self.gate_up_proj_conv(x)
+        gate, up_states = up_states.chunk(2, dim=1)
+        up_states = up_states * self.activation_fn(gate)
+        x = self.down_proj_conv(up_states)
+        x = x.transpose(1, 3)
+        x = torch.reshape(x, (bsz, -1, self.dim))
+        return x
+
+    def forward(self, x):
+        up_states = self.gate_up_proj(x)
+        gate, up_states = up_states.chunk(2, dim=-1)
+        up_states = up_states * self.activation_fn(gate)
+        return self.down_proj(up_states)