Publish MIC usage, slim fixture and reviewed Transformers expectations

.mic-repo.json

@@ -0,0 +1,7 @@
+{
+  "format": "mic-umbrella-v1",
+  "contract_files": [
+    "integration.json",
+    "integrations/transformers/usage.md"
+  ]
+}

README.md

@@ -0,0 +1,37 @@
+---
+tags:
+  - mic
+  - test-artifact
+base_model: ai21labs/Jamba-tiny-dev
+---
+
+# Jamba-tiny-dev-mic
+
+Slim model weights, executable usage and framework expectations in one repository.
+
+- [Contract manifest](integration.json)
+- [Transformers usage](integrations/transformers/usage.md)
+- [Transformers expectations](integrations/transformers/expectations)
+- [Fixture provenance](build.json)
+
+## Contract
+
+### Jamba causal LM
+
+Runnable usage and the framework expectations directory are linked above.
+
+## Slim model
+
+### Slim fixture: Jamba-tiny-dev-mic
+
+A 8.8M-parameter fixture for the [Jamba MIC](https://huggingface.co/huggingface-mic/Jamba-tiny-dev-mic), covering [`ai21labs/Jamba-tiny-dev`](https://huggingface.co/ai21labs/Jamba-tiny-dev). It is a test artifact, not a usable model, tagged `test-artifact` to keep it out of candidate-model search.
+
+- **Fixture key `ce507b862c67`.** It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
+- **Same architecture, smaller config:** `JambaForCausalLM`, shrunk by the generic rule in [`sweep.py`](https://github.com/huggingface/transformers-mic/blob/main/experiments/sweep.py) (num_hidden_layers 5).
+- **Processor files copied byte for byte** from `ai21labs/Jamba-tiny-dev`: `generation_config.json`, `special_tokens_map.json`, `tokenizer.json`, `tokenizer.model`, `tokenizer_config.json`.
+- **Random weights (seed 17).** They passed the input-sensitivity gate: every perturbation of the contract's inputs moved the logits by at least 0.001 (weakest: `input_ids:roll_batch`, 0.764), so no training was needed.
+- **Same tensor names as the real checkpoint.** The real checkpoint has every parameter, and so does the fixture.
+- **Saved in bfloat16**, the dtype `ai21labs/Jamba-tiny-dev` loads in by default.
+- **Finite in float32, bfloat16, and float16** (forward, loss, gradients).
+
+Built by [`build_fixtures.py`](https://github.com/huggingface/transformers-mic/blob/main/example/build_fixtures.py); [`build.json`](build.json) records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with `python example/build_fixtures.py --fixture Jamba-tiny-dev-mic`.

build.json

@@ -0,0 +1,148 @@
+{
+ "fixture": "huggingface-mic/Jamba-tiny-dev-mic",
+ "key": "ce507b862c67085517bdb80699420885f7ba5b0af9beece93aa10ae531c15c80",
+ "key_fields": {
+  "architectures": [
+   "JambaForCausalLM"
+  ],
+  "output_hidden_states": false,
+  "return_dict": true,
+  "chunk_size_feed_forward": 0,
+  "is_encoder_decoder": false,
+  "id2label": {
+   "0": "LABEL_0",
+   "1": "LABEL_1"
+  },
+  "label2id": {
+   "LABEL_0": 0,
+   "LABEL_1": 1
+  },
+  "problem_type": null,
+  "vocab_size": 65536,
+  "tie_word_embeddings": false,
+  "hidden_size": 64,
+  "intermediate_size": 128,
+  "num_hidden_layers": 5,
+  "num_attention_heads": 4,
+  "num_key_value_heads": 2,
+  "hidden_act": "silu",
+  "initializer_range": 0.02,
+  "rms_norm_eps": 1e-06,
+  "use_cache": true,
+  "output_router_logits": false,
+  "router_aux_loss_coef": 0.001,
+  "pad_token_id": 0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "max_position_embeddings": 262144,
+  "attention_dropout": 0.0,
+  "num_experts_per_tok": 2,
+  "num_experts": 4,
+  "expert_layer_period": 2,
+  "expert_layer_offset": 1,
+  "attn_layer_period": 8,
+  "attn_layer_offset": 4,
+  "use_mamba_kernels": true,
+  "mamba_d_state": 16,
+  "mamba_d_conv": 4,
+  "mamba_expand": 2,
+  "mamba_dt_rank": 32,
+  "mamba_conv_bias": true,
+  "mamba_proj_bias": false,
+  "use_mambapy": false,
+  "use_associative_scan": true,
+  "model_type": "jamba",
+  "num_logits_to_keep": 1,
+  "sliding_window": null,
+  "output_attentions": false,
+  "dtype": "bfloat16"
+ },
+ "covers": {
+  "ai21labs/Jamba-tiny-dev": "ed303361004ac875426a61675edecf8e9d976882"
+ },
+ "built_from": {
+  "repo": "ai21labs/Jamba-tiny-dev",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882"
+ },
+ "seed": 17,
+ "dtype": "bfloat16",
+ "builder": "example/build_fixtures.py",
+ "builder_revision": "76cba244631520dc5e504be76e22bb73fcf16b8b",
+ "recipe": {
+  "seed": 17
+ },
+ "shrink": "rule",
+ "model_class": "JambaForCausalLM",
+ "parameters": 8831936,
+ "missing_parameters": [],
+ "checkpoint_inspection": "parameter headers/indexes; bounded binary fallback",
+ "numeric_health_scope": "CPU float32/bfloat16/float16; autocast for low-precision losses",
+ "captured": {
+  "call": "training step",
+  "inputs": [
+   "attention_mask",
+   "input_ids",
+   "labels"
+  ],
+  "class": "JambaForCausalLM"
+ },
+ "weights": "random",
+ "gate": {
+  "random": {
+   "pass": true,
+   "accepted_insensitive": {},
+   "threshold": 0.001,
+   "deltas": {
+    "input_ids:roll_batch": 0.763671875,
+    "input_ids:reverse": 1.17578125,
+    "attention_mask:all_ones": null
+   },
+   "weakest": "input_ids:roll_batch",
+   "no_perturbation_defined": [],
+   "not_a_forward_parameter": []
+  }
+ },
+ "numeric_health": {
+  "float32": {
+   "outputs_finite": true,
+   "loss_finite": true,
+   "gradients_finite": true,
+   "finite": true
+  },
+  "bfloat16": {
+   "outputs_finite": true,
+   "loss_finite": true,
+   "gradients_finite": true,
+   "finite": true
+  },
+  "float16": {
+   "outputs_finite": true,
+   "loss_finite": true,
+   "unscaled_gradients_finite": true,
+   "finite": true
+  }
+ },
+ "copied_unchanged": [
+  "generation_config.json",
+  "special_tokens_map.json",
+  "tokenizer.json",
+  "tokenizer.model",
+  "tokenizer_config.json"
+ ],
+ "environment": {
+  "python": "3.12.12",
+  "machine": "arm64",
+  "torch": "2.14.1",
+  "transformers": "5.22.0.dev0",
+  "tokenizers": "0.23.3"
+ },
+ "sha256": {
+  "config.json": "ce9577f4c98802184f1e5d0b326cea67af4e19840f1d426004811771a54a72a7",
+  "generation_config.json": "0dd850e9c007f3046febe16658a39b6caa96768682f7fc4c356861be6b456216",
+  "model.safetensors": "26d26ea890188b3ce4cf8b1aaa843d141559cedd782f180f0085289b84130f5d",
+  "special_tokens_map.json": "ff56aa3285bed3b10bd29029296c4b290536dc174e54ab9c15e7c4edfa180492",
+  "tokenizer.json": "ce194086b50346971dabd71bba69b9d4c40b3a40c055977dc9a75da9d39080c8",
+  "tokenizer.model": "8b0df4fb43262c452ef37061951a06df4c63ca191d02a60ea08f14428af24376",
+  "tokenizer_config.json": "88bc9a4a7a7dd363d9f559e6e85bdeda3bb49d3e6725252c3b23c3383a0527a0"
+ }
+}

config.json

@@ -0,0 +1,43 @@
+{
+  "architectures": [
+    "JambaForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "attn_layer_offset": 4,
+  "attn_layer_period": 8,
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": 2,
+  "expert_layer_offset": 1,
+  "expert_layer_period": 2,
+  "hidden_act": "silu",
+  "hidden_size": 64,
+  "initializer_range": 0.02,
+  "intermediate_size": 128,
+  "mamba_conv_bias": true,
+  "mamba_d_conv": 4,
+  "mamba_d_state": 16,
+  "mamba_dt_rank": 32,
+  "mamba_expand": 2,
+  "mamba_proj_bias": false,
+  "max_position_embeddings": 262144,
+  "model_type": "jamba",
+  "num_attention_heads": 4,
+  "num_experts": 4,
+  "num_experts_per_tok": 2,
+  "num_hidden_layers": 5,
+  "num_key_value_heads": 2,
+  "num_logits_to_keep": 1,
+  "output_router_logits": false,
+  "pad_token_id": 0,
+  "rms_norm_eps": 1e-06,
+  "router_aux_loss_coef": 0.001,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "5.22.0.dev0",
+  "use_associative_scan": true,
+  "use_cache": true,
+  "use_mamba_kernels": true,
+  "use_mambapy": false,
+  "vocab_size": 65536
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "pad_token_id": 0,
+  "transformers_version": "4.40.2"
+}

integration.json

@@ -0,0 +1,80 @@
+{
+  "mic": "0.3",
+  "name": "Jamba causal LM",
+  "usage": "AutoTokenizer + AutoModelForCausalLM, batched, left padding, labels for one training step",
+  "architecture": "JambaForCausalLM",
+  "checkpoints": {
+    "ai21labs/Jamba-tiny-dev": {
+      "fixture": "huggingface-mic/Jamba-tiny-dev-mic"
+    }
+  },
+  "fixtures": {
+    "huggingface-mic/Jamba-tiny-dev-mic": {
+      "built_from": "ai21labs/Jamba-tiny-dev",
+      "covers": [
+        "ai21labs/Jamba-tiny-dev"
+      ],
+      "build": {
+        "seed": 17
+      }
+    }
+  },
+  "integrations": {
+    "transformers": {
+      "entry_point": "integrations/transformers/usage.md",
+      "format": "runnable-markdown",
+      "requires": [
+        "transformers>=5",
+        "torch"
+      ],
+      "targets": [
+        "fixtures",
+        "checkpoints"
+      ],
+      "variants": {
+        "default": {},
+        "eager": {
+          "attn_implementation": "eager"
+        },
+        "bfloat16": {
+          "dtype": "bfloat16"
+        }
+      },
+      "training": true,
+      "resources": {
+        "fixture": {
+          "memory_gb": 2,
+          "download_mb": 40
+        },
+        "ai21labs/Jamba-tiny-dev": {
+          "memory_gb": 8,
+          "download_mb": 1600
+        }
+      }
+    }
+  },
+  "outputs": {
+    "runtime": {
+      "compare": "exact"
+    },
+    "last_logits": {
+      "compare": "allclose",
+      "atol": 0.0001,
+      "rtol": 0.0001
+    },
+    "next_token_ids": {
+      "compare": "exact"
+    },
+    "training.loss": {
+      "compare": "allclose",
+      "atol": 0.0001,
+      "rtol": 0.0001
+    },
+    "training.gradients": {
+      "compare": "all_true"
+    },
+    "training.updated": {
+      "compare": "true"
+    }
+  }
+}

integrations/transformers/expectations/cpu-x86_64-bf16/Jamba-tiny-dev-mic.bfloat16.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.34,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64-bf16"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -0.166015625,
+    0.09814453125,
+    -0.0267333984375,
+    0.09765625,
+    0.10498046875
+   ],
+   [
+    -0.11962890625,
+    0.09521484375,
+    -0.00897216796875,
+    0.09521484375,
+    0.06396484375
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139739990234375,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64-bf16/Jamba-tiny-dev.bfloat16.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 8.66,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64-bf16"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -7.34375,
+    -7.34375,
+    5.5625,
+    -6.9375,
+    -7.28125
+   ],
+   [
+    -7.78125,
+    -7.9375,
+    4.84375,
+    -7.40625,
+    -7.75
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.08669900894165,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/Jamba-tiny-dev-mic.eager.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.31,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "last_logits": [
+   [
+    -0.16592717170715332,
+    0.09719660133123398,
+    -0.027352558448910713,
+    0.0969439446926117,
+    0.1053500846028328
+   ],
+   [
+    -0.11996722221374512,
+    0.09445894509553909,
+    -0.009807074442505836,
+    0.09533163905143738,
+    0.06285151839256287
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139444351196289,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/Jamba-tiny-dev-mic.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.37,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -0.1659272164106369,
+    0.09719657897949219,
+    -0.02735256403684616,
+    0.09694396704435349,
+    0.10535009205341339
+   ],
+   [
+    -0.11996722221374512,
+    0.09445895254611969,
+    -0.009807072579860687,
+    0.09533164650201797,
+    0.06285154074430466
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139444351196289,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/Jamba-tiny-dev.eager.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 8.93,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "last_logits": [
+   [
+    -7.346984386444092,
+    -7.360681533813477,
+    5.556155204772949,
+    -6.959625244140625,
+    -7.305123805999756
+   ],
+   [
+    -7.784221649169922,
+    -7.924924373626709,
+    4.815289497375488,
+    -7.392331600189209,
+    -7.744963645935059
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.073472499847412,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/Jamba-tiny-dev.json

@@ -0,0 +1,72 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 8.54,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -7.346982955932617,
+    -7.36068058013916,
+    5.556156158447266,
+    -6.959624290466309,
+    -7.305122375488281
+   ],
+   [
+    -7.784221172332764,
+    -7.924923419952393,
+    4.8152899742126465,
+    -7.392330169677734,
+    -7.744962692260742
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.073472499847412,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev-mic.bfloat16.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 8.5,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -0.166015625,
+    0.09814453125,
+    -0.0267333984375,
+    0.09765625,
+    0.10498046875
+   ],
+   [
+    -0.11962890625,
+    0.09521484375,
+    -0.00897216796875,
+    0.09521484375,
+    0.06396484375
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139740943908691,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev-mic.eager.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 8.22,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "last_logits": [
+   [
+    -0.1659272164106369,
+    0.09719658643007278,
+    -0.02735253795981407,
+    0.09694396704435349,
+    0.10535009205341339
+   ],
+   [
+    -0.11996721476316452,
+    0.09445895999670029,
+    -0.00980711355805397,
+    0.09533165395259857,
+    0.06285150349140167
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139444351196289,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev-mic.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "Jamba-tiny-dev-mic",
+  "kind": "fixture",
+  "revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 9.21,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -0.1659272164106369,
+    0.09719659388065338,
+    -0.02735251560807228,
+    0.09694400429725647,
+    0.10535012185573578
+   ],
+   [
+    -0.11996722221374512,
+    0.0944589301943779,
+    -0.009807102382183075,
+    0.09533168375492096,
+    0.06285151839256287
+   ]
+  ],
+  "next_token_ids": [
+   63289,
+   63289
+  ],
+  "training": {
+   "loss": 11.139444351196289,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev.bfloat16.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 9.86,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -7.375,
+    -7.375,
+    5.5625,
+    -7.0,
+    -7.3125
+   ],
+   [
+    -7.78125,
+    -7.9375,
+    4.84375,
+    -7.40625,
+    -7.75
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.0856804847717285,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev.eager.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 10.81,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "last_logits": [
+   [
+    -7.346982955932617,
+    -7.360681056976318,
+    5.556158065795898,
+    -6.959626197814941,
+    -7.305124282836914
+   ],
+   [
+    -7.7842206954956055,
+    -7.924923419952393,
+    4.815289497375488,
+    -7.392331600189209,
+    -7.744964599609375
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.073477745056152,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/Jamba-tiny-dev.json

@@ -0,0 +1,73 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/Jamba-tiny-dev-mic",
+  "contract_name": "Jamba",
+  "contract_revision": "7e840f9d924defee8bbe8fa47d0c69b50b714e0e",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "Jamba-tiny-dev",
+  "kind": "checkpoint",
+  "revision": "ed303361004ac875426a61675edecf8e9d976882",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 10.36,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "last_logits": [
+   [
+    -7.346983432769775,
+    -7.36068058013916,
+    5.556157112121582,
+    -6.959625244140625,
+    -7.305123329162598
+   ],
+   [
+    -7.784223556518555,
+    -7.924923896789551,
+    4.815288543701172,
+    -7.392330169677734,
+    -7.744963645935059
+   ]
+  ],
+  "next_token_ids": [
+   7754,
+   1808
+  ],
+  "training": {
+   "loss": 4.073477745056152,
+   "gradients": {
+    "model.layers.0.mamba.in_proj.weight": true,
+    "model.layers.0.feed_forward.gate_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/usage.md

@@ -0,0 +1,72 @@
+# Jamba: causal language modeling with Transformers
+
+This page is the Transformers contract of the Jamba MIC and also a usage example. [doc-builder](https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md) renders the `runnable` blocks below as documentation, and its pytest plugin runs them as written against `ai21labs/Jamba-tiny-dev`. The MIC runner runs the same blocks against each slim fixture and each checkpoint in [`integration.json`](../../integration.json). Lines marked `# doc-builder: hide` only matter there: they swap in the pinned checkpoint or variant and print the result record.
+
+## Predict the next token
+
+```py runnable:causal_lm
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
+import json  # doc-builder: hide
+import os  # doc-builder: hide
+
+model_id = "ai21labs/Jamba-tiny-dev"
+model_id = os.environ.get("MIC_MODEL", model_id)  # doc-builder: hide
+variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}"))  # doc-builder: hide
+device = "cuda" if torch.cuda.is_available() else "cpu"
+set_seed(0)  # seeds torch, NumPy, and Python: loading, dropout, and data augmentation draw random numbers
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+# Jamba-tiny-dev is stored in bfloat16; load in float32 so a training step does not need loss scaling.
+model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.float32).to(device)
+if variant:  # doc-builder: hide
+    del model  # doc-builder: hide
+    model = AutoModelForCausalLM.from_pretrained(model_id, **{"dtype": torch.float32, **variant}).to(device)  # doc-builder: hide
+
+prompts = [
+    "The capital of France is",
+    "The capital of Germany is",
+]
+inputs = tokenizer(prompts, return_tensors="pt", padding=True).to(device)
+
+with torch.no_grad():
+    logits = model(**inputs).logits
+# Left padding: the last position of each row is the real final token.
+last_logits = logits[:, -1, :]
+next_token_ids = last_logits.argmax(dim=-1)
+print(tokenizer.batch_decode(next_token_ids))
+```
+
+## One training step
+
+```py runnable:causal_lm:2
+labels = inputs["input_ids"].clone()
+labels[inputs["attention_mask"] == 0] = -100
+
+model.train()
+optimizer = torch.optim.SGD([p for p in model.parameters() if p.requires_grad], lr=0.01)
+loss = model(**inputs, labels=labels).loss
+loss.backward()
+before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad}  # doc-builder: hide
+optimizer.step()
+print(f"loss: {loss.item():.4f}")
+
+if os.environ.get("MIC_RUNNER"):  # doc-builder: hide
+    named = dict(model.named_parameters())
+    checked = ("model.layers.0.mamba.in_proj.weight", "model.layers.0.feed_forward.gate_proj.weight")
+    result = {
+        "runtime": {
+            "device": model.device.type,
+            "dtype": str(model.dtype).removeprefix("torch."),
+            "attn_implementation": model.config._attn_implementation,
+        },
+        "last_logits": last_logits[:, :5].float().tolist(),
+        "next_token_ids": next_token_ids.tolist(),
+        "training": {
+            "loss": loss.item(),
+            "gradients": {n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all()) for n in checked},
+            "updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before),
+        },
+    }
+    print(json.dumps(result))
+```

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26d26ea890188b3ce4cf8b1aaa843d141559cedd782f180f0085289b84130f5d
+size 17674944

special_tokens_map.json

@@ -0,0 +1,48 @@
+{
+  "additional_special_tokens": [
+    "<|eom|>",
+    "<|bom|>",
+    "<|system|>",
+    "<|user|>",
+    "<|assistant|>",
+    "<|tool|>",
+    "<documents>",
+    "</documents>",
+    "<tool_definitions>",
+    "</tool_definitions>",
+    "<active_output_modes>",
+    "</active_output_modes>",
+    "<citations>",
+    "</citations>",
+    "<tool_calls>",
+    "</tool_calls>"
+  ],
+  "bos_token": {
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|pad|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<|unk|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b0df4fb43262c452ef37061951a06df4c63ca191d02a60ea08f14428af24376
+size 1124714

tokenizer_config.json

@@ -0,0 +1,196 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": null,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<|pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<|startoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<|unk|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "518": {
+      "content": "<|eom|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "519": {
+      "content": "<|bom|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "520": {
+      "content": "<|system|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "521": {
+      "content": "<|user|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "522": {
+      "content": "<|assistant|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "523": {
+      "content": "<|tool|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "524": {
+      "content": "<documents>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "525": {
+      "content": "</documents>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "526": {
+      "content": "<tool_definitions>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "527": {
+      "content": "</tool_definitions>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "528": {
+      "content": "<active_output_modes>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "529": {
+      "content": "</active_output_modes>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "530": {
+      "content": "<citations>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "531": {
+      "content": "</citations>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "532": {
+      "content": "<tool_calls>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "533": {
+      "content": "</tool_calls>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|eom|>",
+    "<|bom|>",
+    "<|system|>",
+    "<|user|>",
+    "<|assistant|>",
+    "<|tool|>",
+    "<documents>",
+    "</documents>",
+    "<tool_definitions>",
+    "</tool_definitions>",
+    "<active_output_modes>",
+    "</active_output_modes>",
+    "<citations>",
+    "</citations>",
+    "<tool_calls>",
+    "</tool_calls>"
+  ],
+  "bos_token": "<|startoftext|>",
+  "chat_template": "{# Variables #}\n{% set ns = namespace(message_count=0, is_last_checked_defined=False) %}\n{##}\n{% set bom_str = bom_str or \"<|bom|>\" %}\n{% set eom_str = eom_str or \"<|eom|>\" %}\n{% set default_system_message = \"\" %}\n{##}\n{% set documents_prefix = \"<documents>\" %}\n{% set documents_suffix = \"</documents>\" %}\n{% set tool_definitions_prefix = \"<tool_definitions>\" %}\n{% set tool_definitions_suffix = \"</tool_definitions>\" %}\n{% set active_modes_prefix = \"<active_output_modes>\" %}\n{% set active_modes_suffix = \"</active_output_modes>\" %}\n{##}\n{% set tool_calls_prefix = \"<tool_calls>\" %}\n{% set tool_calls_suffix = \"</tool_calls>\" %}\n{% set citations_prefix = \"<citations>\" %}\n{% set citations_suffix = \"</citations>\" %}\n{##}\n{% if add_generation_prompt is not defined %}\n  {% set add_generation_prompt = True %}\n{% endif %}\n{% set role_to_predict = role_to_predict or \"assistant\" %}\n{% if messages|length > 0 and messages[0].role == \"system\" %}\n  {% set system_message = messages[0].content %}\n  {% set loop_messages = messages[1:] %}\n{% else %}\n  {% set system_message = default_system_message %}\n  {% set loop_messages = messages %}\n{% endif %}\n{##}\n{##}\n{# Macros #}\n{% macro handle_tool_definitions(tools) %}\n  {{- tool_definitions_prefix -}}\n  {{- \"\\n# Tools\" -}}\n  {{- \"\\n\\n## Functions\" -}}\n  {% for tool in tools %}\n    {% set _ = is_param_set(tool, field=\"type\") %}\n    {% set is_tool_type_set = ns.is_last_checked_defined %}\n    {% if is_tool_type_set %}\n      {% if tool.type == \"function\" %}\n        {% set tool = tool.function %}\n      {% else %}\n        {{ raise_exception(\"Currently, the only supported tool type is `function`\") }}\n      {% endif %}\n    {% endif %}\n    {{- \"\\n\\n\" + (tool|tojson(indent=2)) -}}\n  {% endfor %}\n  {{- \"\\n\" + tool_definitions_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_first_system_message(system_message, tools) %}\n  {{- bom_str + handle_role(\"system\") -}}\n  {% set _ = is_param_set(system_message) %}\n  {% set is_system_message_set = ns.is_last_checked_defined %}\n  {% if is_system_message_set %}\n    {{- system_message -}}\n  {% endif %}\n  {% set _ = is_param_set(tools, is_list=True) %}\n  {% set is_tools_set = ns.is_last_checked_defined %}\n  {% if is_tools_set %}\n    {% if system_message %}\n      {{- \"\\n\\n\" -}}\n    {% endif %}\n    {{- handle_tool_definitions(tools) -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endmacro %}\n{##}\n{% macro handle_tool_calls(tool_calls) %}\n  {{- tool_calls_prefix + \"[\\n\" -}}\n  {% for tool_call in tool_calls %}\n    {% set _ = is_param_set(tool_call, field=\"function\") %}\n    {% set is_tool_call_function_set = ns.is_last_checked_defined %}\n    {% if is_tool_call_function_set %}\n      {%- set tool_call = tool_call.function %}\n    {%- endif %}\n    {% set arguments = tool_call.arguments %}\n    {% if arguments is not string %}\n      {%- set arguments = arguments|tojson -%}\n    {%- endif %}\n    {{ \"{\\\"name\\\": \\\"\" + tool_call.name + \"\\\", \\\"arguments\\\": \" + arguments + \"}\" -}}\n    {% if not loop.last %}\n      {{- \",\" }}\n    {% endif %}\n  {% endfor %}\n  {{- \"\\n]\" + tool_calls_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_documents(documents) %}\n  {{- documents_prefix -}}\n  {{- \"\\n# Documents\" -}}\n  {{- \"\\n\\nYou can use the following documents for reference:\" -}}\n  {% for doc in documents %}\n    {{- \"\\n\\n## Document ID: \" + loop.index0|string -}}\n    {% set _ = is_param_set(doc, field=\"title\") %}\n    {% set is_doc_title_set = ns.is_last_checked_defined %}\n    {% if is_doc_title_set %}\n      {{- \"\\nTitle: \" + doc.title -}}\n    {% endif %}\n    {% for key, value in doc.items() %}\n      {% if key not in [\"title\", \"text\"] %}\n        {{- \"\\n\" + key|title + \": \" + value|string -}}\n      {% endif %}\n    {% endfor %}\n    {{- \"\\nText: \" + doc.text -}}\n  {% endfor %}\n  {{- \"\\n\" + documents_suffix -}}\n{% endmacro %}\n{##}\n{% macro handle_knobs(knobs) %}\n  {{- active_modes_prefix -}}\n  {{- \"\\n# Active Modes\" -}}\n  {{ \"\\n\\nThe following modes configure the format or style of your responses. You should adhere to all currently\" -}}\n  {{ \" active modes simultaneously.\" -}}\n  {% if knobs.citation_mode == \"fast\" %}\n    {{- \"\\n\\n## Citation Mode\" -}}\n    {{- \"\\n\\nProvide a list of references only for the documents you base your response on. Format your response\" -}}\n    {{ \" with the original answer followed by a citation section. Use this template:\" -}}\n    {{ \" `{answer}\" + citations_prefix + \"DOCUMENT_IDS\" + citations_suffix + \"`, where DOCUMENT_IDS are the relevant document numbers\" -}}\n    {{ \" (e.g. [2, 5, 9]), or [] if the answer cannot be supported by the provided documents.\" -}}\n  {% endif %}\n  {% if knobs.response_format == \"json_object\" %}\n    {{- \"\\n\\n## JSON Mode\" -}}\n    {{ \"\\n\\nProvide your response in JSON format. Adhere strictly to any schema given by the user.\" -}}\n    {{ \" If an appropriate JSON format exists, use it without modification.\" -}}\n  {% endif %}\n  {{- \"\\n\" + active_modes_suffix -}}\n{% endmacro %}\n{##}\n{% macro get_last_user_index(messages) %}\n  {% set ns.last_user_index = 0 %}\n  {% for message in messages %}\n    {% if message.role == 'user' %}\n      {% set ns.last_user_index = loop.index0 %}\n    {% endif %}\n  {% endfor %}\n  {{- ns.last_user_index -}}\n{% endmacro %}\n{##}\n{% macro handle_last_system_message(documents, knobs, use_documents, use_knobs) %}\n  {{- bom_str + handle_role(\"system\") -}}\n  {% set macros_to_call = [] %}\n  {% set params_for_macros = [] %}\n  {% if use_documents %}\n    {% set macros_to_call = macros_to_call + [handle_documents] %}\n    {% set params_for_macros = params_for_macros + [[documents]] %}\n  {% endif %}\n  {% if use_knobs %}\n    {% set macros_to_call = macros_to_call + [handle_knobs] %}\n    {% set params_for_macros = params_for_macros + [[knobs]] %}\n  {% endif %}\n  {% for i in range(macros_to_call|length) %}\n    {% if i > 0 %}\n      {{- \"\\n\\n\" -}}\n    {% endif %}\n    {{- macros_to_call[i](*params_for_macros[i]) -}}\n  {% endfor %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endmacro %}\n{##}\n{% macro handle_role(role, add_space=True) %}\n  {{- \"<|\" + role + \"|>\" -}}\n  {% if add_space %}\n    {{- \" \" -}}\n  {% endif %}\n{% endmacro %}\n{##}\n{% macro is_param_set(param, field=none, is_list=False) %}\n  {% if field is not none %}\n    {% if field in param %}\n      {% set param = param[field] %}\n    {% else %}\n      {% set param = none %}\n    {% endif %}\n  {% endif %}\n  {% set is_defined = param is defined and param is not none %}\n  {% if is_list %}\n    {% set ns.is_last_checked_defined = is_defined and param|length > 0 %}\n  {% else %}\n    {% set ns.is_last_checked_defined = is_defined %}\n  {% endif %}\n{% endmacro %}\n{##}\n{##}\n{# Template #}\n{{- \"<|startoftext|>\" -}}\n{% set _ = is_param_set(system_message) %}\n{% set is_system_message_set = ns.is_last_checked_defined %}\n{% set _ = is_param_set(tools, is_list=True) %}\n{% set is_tools_set = ns.is_last_checked_defined %}\n{% set has_system_message = (is_system_message_set or is_tools_set) %}\n{% if has_system_message %}\n  {{- handle_first_system_message(system_message, tools) -}}\n{% endif %}\n{% set last_user_index = get_last_user_index(loop_messages)|int %}\n{% for message in loop_messages %}\n  {% if loop.index0 == last_user_index %}\n    {% set _ = is_param_set(documents, is_list=True) %}\n    {% set use_documents = ns.is_last_checked_defined %}\n    {% set _ = is_param_set(knobs) %}\n    {% set use_knobs = ns.is_last_checked_defined and knobs.is_set %}\n    {% set add_last_system_message = use_documents or use_knobs %}\n    {% if add_last_system_message %}\n      {% if ns.message_count > 0 %}\n        {{- eom_str -}}\n      {% endif %}\n      {{- handle_last_system_message(documents, knobs, use_documents, use_knobs) -}}\n    {% endif %}\n  {% endif %}\n  {% set role = message.role %}\n  {% set _ = is_param_set(message, field=\"name\") %}\n  {% set is_message_name_set = ns.is_last_checked_defined %}\n  {% if is_message_name_set %}\n    {% set message_prefix = handle_role(role) + \"(\" + message.name + \")\" %}\n  {% else %}\n    {% set message_prefix = handle_role(role) %}\n  {% endif %}\n  {% set content = (message.content or \"\") %}\n  {% if content is not string %}\n    {% set content = content|tojson %}\n  {% endif %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n  {{- bom_str + message_prefix + content -}}\n  {% set _ = is_param_set(message, field=\"tool_calls\", is_list=True) %}\n  {% set is_tool_calls_set = ns.is_last_checked_defined %}\n  {% if role == \"assistant\" and is_tool_calls_set %}\n    {{- handle_tool_calls(message.tool_calls) -}}\n  {% endif %}\n  {% set _ = is_param_set(message, field=\"citations\", is_list=True) %}\n  {% set is_citations_set = ns.is_last_checked_defined %}\n  {% if role == \"assistant\" and is_citations_set %}\n    {{- citations_prefix + message.citations|map(attribute=\"document_id\")|list|string + citations_suffix -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% endfor %}\n{% if add_generation_prompt %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n  {{- bom_str + handle_role(role_to_predict, add_space=False) -}}\n  {% set _ = is_param_set(generation_preamble) %}\n  {% set is_generation_preamble_set = ns.is_last_checked_defined %}\n  {% if is_generation_preamble_set and generation_preamble.strip() != \"\" %}\n    {{- \" \" + generation_preamble -}}\n  {% endif %}\n  {% set ns.message_count = ns.message_count + 1 %}\n{% else %}\n  {% if ns.message_count > 0 %}\n    {{- eom_str -}}\n  {% endif %}\n{% endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<|pad|>",
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<|unk|>",
+  "use_default_system_prompt": false
+}