Fix use of Falcon language model for text generation (#3592)

* fix falcon

* change model parameters to not use additional requirements

* remove additional test dependencies

shap/models/_teacher_forcing.py

@@ -1,3 +1,5 @@
+import inspect
+
 import numpy as np
 import scipy.special
 
@@ -282,6 +284,8 @@ def model_inference(self, inputs, output_ids):
                     inputs["position_ids"] = (inputs["attention_mask"].long().cumsum(-1) - 1)
                     inputs["position_ids"].masked_fill_(inputs["attention_mask"] == 0, 0)
                     # model inference
+                    expected_parameters = list(inspect.signature(self.similarity_model.forward).parameters)
+                    inputs = {k: v for k, v in inputs.items() if k in expected_parameters}
                     outputs = self.similarity_model(**inputs, return_dict=True)
                 logits = outputs.logits.detach().cpu().numpy().astype('float64')
         elif self.similarity_model_type == "tf":

tests/models/test_teacher_forcing_logits.py

@@ -6,6 +6,39 @@
 import shap
 
 
+def test_falcon():
+    transformers = pytest.importorskip("transformers")
+    requests = pytest.importorskip("requests")
+    name = "fxmarty/really-tiny-falcon-testing"
+    try:
+        tokenizer = transformers.AutoTokenizer.from_pretrained(name)
+        model = transformers.AutoModelForCausalLM.from_pretrained(
+            name, trust_remote_code=True, load_in_8bit=False, low_cpu_mem_usage=False
+        )
+    except requests.exceptions.RequestException:
+        pytest.xfail(reason="Connection error to transformers model")
+
+    model = model.eval()
+
+
+    s = ["I enjoy walking with my cute dog"]
+    gen_dict = dict(
+        max_new_tokens=100,
+        num_beams=5,
+        renormalize_logits=True,
+        no_repeat_ngram_size=8,
+    )
+
+
+    model.config.task_specific_params = dict()
+    model.config.task_specific_params["text-generation"] = gen_dict
+    shap_model = shap.models.TeacherForcing(model, tokenizer)
+
+    explainer = shap.Explainer(shap_model, tokenizer)
+    shap_values = explainer(s)
+    assert not np.isnan(np.sum(shap_values.values))
+
+
 def test_method_get_teacher_forced_logits_for_encoder_decoder_model():
     """Tests if get_teacher_forced_logits() works for encoder-decoder models."""
     transformers = pytest.importorskip("transformers")