mmangkad/sarvam-105b-bf16

Sarvam-105B — BF16

A bfloat16 conversion of `sarvamai/sarvam-105b`, which is distributed in float32.

This conversion is bit-exact, not lossy. Every one of the 106,031,767,424 parameters in the original was verified to already be exactly bf16-representable — all 16 low mantissa bits were zero across the entire checkpoint. The model was evidently trained in bf16 and stored widened to fp32, so narrowing it back discards only known-zero bits. Upcasting these weights to fp32 reproduces the original file contents bitwise.

Verification

Each converted tensor was upcast back to fp32 and compared against the source as raw int32 bit patterns:

shards      : 85/85
tensors     : 12,289/12,289
parameters  : 106,031,767,424
mismatches  : 0
NaN / Inf   : 0 / 0

Only two files differ from the original repo: config.json ("dtype": "float32" → "bfloat16") and model.safetensors.index.json (total_size halved). All other files, including the tokenizer and the modeling code, are byte-identical.

Usage

Identical to the original — the architecture, tokenizer and remote code are unchanged:

from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("mmangkad/sarvam-105b-bf16", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    "mmangkad/sarvam-105b-bf16", dtype="bfloat16", device_map="auto", trust_remote_code=True,
)

All credit for the model itself goes to Sarvam AI. Licensed apache-2.0, as the original. The original model card follows unmodified.

Want a smaller model? Download Sarvam-30B!

Index

1. 1.Introduction
2. 2.Architecture
3. 3.Benchmarks
4. 4.Knowledge & Coding
5. 5.Reasoning & Math
6. 6.Agentic
7. 7.Inference
8. 8.Hugging Face
9. 9.vLLM
10. 10.SGLang
11. 11.Footnote
12. 12.Citation

Introduction

Sarvam-105B is an advanced Mixture-of-Experts (MoE) model with 10.3B active parameters, designed for superior performance across a wide range of complex tasks. It is highly optimized for complex reasoning, with particular strength in agentic tasks, mathematics, and coding.

Sarvam-105B is a top-tier performer, consistently matching or surpassing several major closed-source models and staying within a narrow margin of frontier models across diverse reasoning and agentic benchmarks. It demonstrates exceptional agentic and reasoning capabilities in real-world applications such as web search and technical troubleshooting.

A major focus during training was the Indian context and languages, resulting in state-of-the-art performance across 22 Indian languages for its model size.

Sarvam-105B is open-sourced under the Apache License. For more details, see our blog.

Architecture

The 105B model adopts an MLA-style attention stack with decoupled QK head dimensions (q_head_dim=192 split into RoPE and noPE components, v_head_dim=128) and a large headdim of 576, enabling higher representational bandwidth per head while keeping the hidden size at 4096. This approach improves attention expressivity and long-context extrapolation (via YaRN scaling with a factor of 40 and 128K context). It has an `intermediatesize (16384) and moeintermediatesize` (2048), combined with top-8 routing over 128 experts, which increases per-token active capacity while keeping activation cost manageable. The model has one shared expert, a routed scaling factor of 2.5, and auxiliary-loss-free router balancing.

Benchmarks

<details> <summary>Knowledge & Coding</summary>

</details>

<details> <summary>Reasoning & Math</summary>

</details>

<details> <summary>Agentic</summary>

See footnote for evaluation details.

</details>

Inference

<details> <summary>Huggingface</summary>

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig

model_name = "sarvamai/sarvam-105b"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, device_map="auto")

def generate_text(
    prompt: str,
    max_new_tokens: int = 2048,
    temperature: float = 0.8,
    top_p: float = 0.95,
    repetition_penalty: float = 1.0,
) -> None:
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda:0")

    generation_config = GenerationConfig(
        max_new_tokens=max_new_tokens,
        repetition_penalty=repetition_penalty,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
    )

    with torch.no_grad():
        output_ids = model.generate(
            input_ids=inputs["input_ids"],
            attention_mask=inputs["attention_mask"],
            generation_config=generation_config,
        )
    return tokenizer.decode(output_ids[0], skip_special_tokens=True)

prompts = [
    "Which country won the FIFA World Cup in 2012?",
]

for prompt in prompts:
    templated_prompt = tokenizer.apply_chat_template(
      [{"role": "user", "content": prompt}],
      tokenize=False,
      add_generation_prompt=True,
      enable_thinking=True
    )
    output = generate_text(templated_prompt, max_new_tokens=512)
    print("Prompt: ", prompt)
    print("Generated text: ", output)
    print("=" * 100)

</details>

<details> <summary>SGLang</summary>

Install latest SGLang from source

git clone https://github.com/sgl-project/sglang.git
cd sglang
pip install -e "python[all]"

Instantiate model and Run

import sglang as sgl
from transformers import AutoTokenizer

model_path = "sarvamai/sarvam-105b"
engine = sgl.Engine(
    model_path=model_path,
    tp_size=4,
    mem_fraction_static=0.70,
    trust_remote_code=True,
    dtype="bfloat16",
    moe_runner_backend="flashinfer_cutedsl",
    prefill_attention_backend="fa3",
    decode_attention_backend="flashmla",
    disable_radix_cache=False,
)

sampling_params = {
    "temperature": 0.8,
    "max_new_tokens": 2048,
    "repetition_penalty": 1.0,
}

prompts = [
    "Which band released the album Dark Side of the Moon in 1973?",
]

outputs = engine.generate([
    tokenizer.apply_chat_template([
        {"role": "user", "content": prompt}],
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=True)
        for prompt in prompts], 
    sampling_params)
for p, o in zip(prompts, outputs):
    print("Prompt: ", p)
    print("Generated text: ", o['text'])
    print("=" * 100)

</details>

<details> <summary>vLLM</summary>

Note: currently a PR is open for native support for the Sarvam models in vLLM (link). Therefore, we have 2 options here.

Option 1: install from source (hard)

• —Use the custom fork here: link
• —Follow the instructions here to install from source: link

Option 2: hot-patch (easy)

• —Run hotpatch_vllm.py
• —This will do the following:
• —install vllm=0.15.0
• —add 2 model entries to registry.py
• —download the model executors for sarvam-105b and sarvam-30b

Once this is done, you can run vLLM as usual

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_path = "sarvamai/sarvam-105b"
tokenizer = AutoTokenizer.from_pretrained(model_path)
llm = LLM(model=model_path, 
            trust_remote_code=True, 
            max_model_len=2048, 
            tensor_parallel_size=8, 
            max_num_seqs=16,
        )
sampling_params = SamplingParams(
                    temperature=0.8, 
                    max_tokens=2048, 
                    repetition_penalty=1.0,
                    spaces_between_special_tokens=True
                )

prompts = [
    "Which artist painted The Persistence of Memory (the melting clocks)?",
]

outputs = llm.generate([
    tokenizer.apply_chat_template([
        {"role": "user", "content": prompt}],
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=True)
        for prompt in prompts], 
    sampling_params)
for p, o in zip(prompts, outputs):
    print("Prompt: ", p)
    print("Generated text: ", o.outputs[0].text)
    print("=" * 100)

</details>

Footnote

• —General settings: All benchmarks are evaluated with a maximum context length of 65,536 tokens.
• —Reasoning & Math benchmarks (Math500, MMLU, MMLU Pro, GPQA Diamond, AIME 25, Beyond AIME, HMMT): Evaluated with temperature=1.0, top_p=1.0, max_new_tokens=65536.
• —Coding & Knowledge benchmarks (Live Code Bench v6, Arena Hard v2, IF Eval): Evaluated with temperature=1.0, top_p=1.0, max_new_tokens=65536.
• —Writing Bench: Responses generated using official Writing-Bench parameters: temperature=0.7, top_p=0.8, top_k=20, max_length=16000. Scoring performed using the official Writing-Bench critic model with: temperature=1.0, top_p=0.95, max_length=2048.
• —Agentic benchmarks (BrowseComp, SWE Bench Verified, τ² Bench): Evaluated with temperature=0.5, top_p=1.0, max_new_tokens=32768.

Citation

@misc{sarvam_sovereign_models,
  title        = {Introducing Sarvam's Sovereign Models},
  author       = {{Sarvam Foundation Models Team}},
  year         = {2026},
  howpublished = {\url{https://www.sarvam.ai/blogs/sarvam-30b-105b}},
  note         = {Accessed: 2026-03-03}
}