AC2513 commited on
Commit
e6d4031
·
1 Parent(s): e15287f

finished model

Browse files
Files changed (1) hide show
  1. Documentation/Technical_Report.md +52 -7
Documentation/Technical_Report.md CHANGED
@@ -8,17 +8,62 @@ Basing off of the Geema 3 technical report, I will analyze and compare different
8
 
9
  The Gemma 3 family consists of four model sizes, each with increasing capabilities and resource requirements:
10
 
11
- | Model | Parameters | Vision Encoder | Total Size | Context Length | Key Capabilities |
12
- |-------|------------|---------------|------------|----------------|------------------|
13
- | Gemma 3-1B | 698M language + 302M embeddings | None | 1B | 32K tokens | Basic text generation; no vision capabilities, extrmely lightweight |
14
- | Gemma 3-4B | 3.2B language + 675M embeddings + 417M vision | SigLIP | 4.3B | 128K tokens | Multimodal with good balance of performance and efficiency |
15
- | Gemma 3-12B | 10.8B language + 1B embeddings + 417M vision | SigLIP | 12.2B | 128K tokens | Strong performance across all tasks with reasonable resource needs |
16
- | Gemma 3-27B | 25.6B language + 1.4B embeddings + 417M vision | SigLIP | 27.4B | 128K tokens | Best performance; comparable to Gemini 1.5 Pro on benchmarks |
 
 
17
 
18
  ### Performance Analysis
19
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20
  ### HuggingFace Spaces Resources
21
 
 
 
 
 
 
 
22
  ## User Research
23
 
24
  ### User Profiles
@@ -29,4 +74,4 @@ The Gemma 3 family consists of four model sizes, each with increasing capabiliti
29
 
30
  ### Technical Stack
31
 
32
- ### UI Interface
 
8
 
9
  The Gemma 3 family consists of four model sizes, each with increasing capabilities and resource requirements:
10
 
11
+ | Model | Parameters | Vision Encoder | Total Size | Context Length | Key Capabilities |
12
+ | ----------- | -------------------------------------------------- | -------------- | ---------- | -------------- | ------------------------------------------------------------------- |
13
+ | Gemma 3-1B | 698M non-embedding + 302M embedding | None | 1B | 32K tokens | Basic text generation; no vision capabilities, extrmely lightweight |
14
+ | Gemma 3-4B | 3.2B non-embedding + 675M embedding + 417M vision | SigLIP | 4.3B | 128K tokens | Multimodal with good balance of performance and efficiency |
15
+ | Gemma 3-12B | 10.8B non-embedding + 1B embedding + 417M vision | SigLIP | 12.2B | 128K tokens | Strong performance across all tasks with reasonable resource needs |
16
+ | Gemma 3-27B | 25.6B non-embedding + 1.4B embedding + 417M vision | SigLIP | 27.4B | 128K tokens | Best performance; comparable to Gemini 1.5 Pro on benchmarks |
17
+
18
+ From this chart, the 1B model does not support vision encoding, thus it is limited to pure text-based tasks. As such, in order to fulloy demonstrate the capability demonstrated by the Gemma models, I will be moving forward with the other three models.
19
 
20
  ### Performance Analysis
21
 
22
+ #### Coding Performance
23
+
24
+ | Model | HumanEval | MBPP | LiveCodeBench |
25
+ | ----- | --------- | ----- | ------------- |
26
+ | 1B | 41.5% | 35.2% | 5.0% |
27
+ | 4B | 71.3% | 63.2% | 23.0% |
28
+ | 12B | 85.4% | 73.0% | 32.0% |
29
+ | 27B | 87.8% | 74.4% | 39.0% |
30
+
31
+ The 12B and 27B models show strong coding capabilities, with 27B achieving the highest accuracy across all code-focused benchmarks. These results indicate that both are well-suited for code generation, debugging assistance, and live programming support. The 4B model, while not at the top, still demonstrates reliable code performance and may serve well in resource-constrained environments.
32
+
33
+ #### Research Capabilities
34
+
35
+ | Model | MMLU | MATH | GSM8K | GPQA Diamond |
36
+ | ----- | ----- | ----- | ----- | ------------ |
37
+ | 1B | 38.8% | 48.0% | 62.8% | 19.2% |
38
+ | 4B | 58.1% | 75.6% | 89.2% | 30.8% |
39
+ | 12B | 71.9% | 83.8% | 94.4% | 40.9% |
40
+ | 27B | 76.9% | 89.0% | 95.9% | 42.4% |
41
+
42
+ In tasks requiring factual recall, mathematical reasoning, and complex QA, performance improves significantly with scale. The 27B model again leads across all metrics.
43
+
44
+ ### Hardware Requirements
45
+
46
+ | Model | bf16 | Int4 | Int4 (Blocks=32) | SFP8 |
47
+ | ------- | ---- | ---- | ---------------- | ---- |
48
+ | **1B** | 2.0 | 0.5 | 0.7 | 1.0 |
49
+ | +KV | 2.9 | 1.4 | 1.6 | 1.9 |
50
+ | **4B** | 8.0 | 2.6 | 2.9 | 4.4 |
51
+ | +KV | 12.7 | 7.3 | 7.6 | 9.1 |
52
+ | **12B** | 24.0 | 6.6 | 7.1 | 12.4 |
53
+ | +KV | 38.9 | 21.5 | 22.0 | 27.3 |
54
+ | **27B** | 54.0 | 14.1 | 15.3 | 27.4 |
55
+ | +KV | 72.7 | 32.8 | 34.0 | 46.1 |
56
+
57
+ These are the required VRAM sizes (in GB) for running the Gemma 3 models under different precision formats and with or without Key-Value (KV) caching. For this application, I will be considering Key-Value pairing as essential for optimizing inference latency and enabling efficient long-context performance; since those are important factors in providing a smooth and responsive user experience in real-time chat scenarios.
58
+
59
  ### HuggingFace Spaces Resources
60
 
61
+ Since I am deploying on HuggingFace Spaces, I will be utilizing their **ZeroGPU**, which offers access to high-performance virtual GPUs, particularly the Nvidia H200 with 70GB of VRAM. This configuration provides ample memory and compute power to run large-scale language models with Key-Value (KV) caching, long context windows, and multimodal inference, all with low latency and high throughput.
62
+
63
+ ### Final Choice
64
+
65
+ For my final choice, in order to fully demonstrate the capabilities of the Gemma 3 family, I have selected the Gemma 3-27B model with Key-Value caching enabled. This setup leverages the full compute and memory bandwidth of the NVIDIA H200 (70GB VRAM) provided by HuggingFace's ZeroGPU environment. Overall, this configuration strikes a strong balance between maximum model capability and inference efficiency, ensuring that the demo remains smooth, accurate, and production-ready — even when scaling to complex or multimodal inputs.
66
+
67
  ## User Research
68
 
69
  ### User Profiles
 
74
 
75
  ### Technical Stack
76
 
77
+ ### UI Interface