Chapter 11.2 - Generating Text Batches

---

Overview

every epoch can contain many batches. every batch can contain many training sets. simple analogy is that 1 batch= the weights changed once 1 epoch= the entire dataset has been read through once

Suppose we have:

• 1000 Training Sets
• Batch Size = 50

2. Number of batches:

   • 1000 ÷ 50 = 20 batches

3. Each batch processes 50 training sets and performs 1 weight update.

4. Therefore:

   • 20 batches = 20 weight updates

5. After all 20 batches are processed, the model has seen all 1000 training sets once. This is called 1 Epoch.

6. If we train for 10 epochs:

   • 20 batches/epoch × 10 epochs = 200 weight updates

Epoch 1 

Batch 1 (Examples 1–50) 
↓ Update weights

Batch 2 (Examples 51–100)
↓ Update weights... 

Batch 20 (Examples 951–1000) ↓ 
Update weights✅ 

Entire dataset has now been seen once 
= 1 Epoch Completed (20 times weight changed)

---

sentance = "hi my name is harshu who really loves my family"

Batch 1

Context Seen	Output	Probability of Output	Target	Probability of Target	Full Probability Matrix Strip [hi, my, name, is, harshu, who, really, loves, family]
hi	hi	0.24	my	0.13	[0.24, 0.13, 0.11, 0.09, 0.12, 0.10, 0.08, 0.07, 0.06]
hi my	really	0.18	name	0.10	[0.09, 0.12, 0.10, 0.11, 0.13, 0.14, 0.18, 0.07, 0.06]
hi my name	my	0.21	is	0.12	[0.08, 0.21, 0.09, 0.12, 0.13, 0.11, 0.10, 0.09, 0.07]
hi my name is	harshu	0.29	harshu	0.29	[0.05, 0.06, 0.07, 0.09, 0.29, 0.16, 0.12, 0.09, 0.07]
my	my	0.26	name	0.12	[0.08, 0.26, 0.12, 0.10, 0.08, 0.14, 0.09, 0.07, 0.06]
my name	really	0.20	is	0.09	[0.07, 0.11, 0.15, 0.09, 0.14, 0.13, 0.20, 0.06, 0.05]
my name is	who	0.22	harshu	0.11	[0.06, 0.09, 0.10, 0.13, 0.11, 0.22, 0.12, 0.10, 0.07]
my name is harshu	who	0.31	who	0.31	[0.04, 0.06, 0.08, 0.07, 0.09, 0.31, 0.14, 0.13, 0.08]

---

Batch 2

Context Seen	Output	Probability of Output	Target	Probability of Target	Full Probability Matrix Strip [hi, my, name, is, harshu, who, really, loves, family]
name	name	0.24	is	0.13	[0.06, 0.12, 0.24, 0.13, 0.09, 0.14, 0.10, 0.07, 0.05]
name is	who	0.21	harshu	0.12	[0.05, 0.08, 0.10, 0.18, 0.12, 0.21, 0.11, 0.09, 0.06]
name is harshu	really	0.20	who	0.14	[0.05, 0.07, 0.08, 0.09, 0.16, 0.14, 0.20, 0.13, 0.08]
name is harshu who	really	0.30	really	0.30	[0.04, 0.05, 0.06, 0.06, 0.09, 0.12, 0.30, 0.18, 0.10]
is	is	0.25	harshu	0.13	[0.05, 0.09, 0.08, 0.25, 0.13, 0.14, 0.10, 0.09, 0.07]
is harshu	who	0.29	who	0.29	[0.04, 0.06, 0.06, 0.08, 0.11, 0.29, 0.15, 0.13, 0.08]
is harshu who	loves	0.28	really	0.16	[0.04, 0.06, 0.06, 0.07, 0.08, 0.17, 0.16, 0.28, 0.08]
is harshu who really	loves	0.35	loves	0.35	[0.03, 0.04, 0.04, 0.05, 0.05, 0.09, 0.15, 0.35, 0.20]

---

Batch 3

Context Seen	Output	Probability of Output	Target	Probability of Target	Full Probability Matrix Strip [hi, my, name, is, harshu, who, really, loves, family]
harshu	harshu	0.24	who	0.12	[0.05, 0.09, 0.08, 0.08, 0.24, 0.12, 0.13, 0.15, 0.06]
harshu who	really	0.28	really	0.28	[0.04, 0.06, 0.05, 0.05, 0.09, 0.16, 0.28, 0.18, 0.09]
harshu who really	loves	0.34	loves	0.34	[0.03, 0.04, 0.04, 0.05, 0.05, 0.10, 0.16, 0.34, 0.19]
harshu who really loves	family	0.33	my	0.18	[0.03, 0.18, 0.04, 0.05, 0.05, 0.06, 0.08, 0.18, 0.33]
who	who	0.23	really	0.16	[0.04, 0.09, 0.08, 0.08, 0.09, 0.23, 0.16, 0.14, 0.09]
who really	loves	0.31	loves	0.31	[0.03, 0.05, 0.05, 0.05, 0.06, 0.12, 0.18, 0.31, 0.15]
who really loves	my	0.39	my	0.39	[0.03, 0.39, 0.04, 0.04, 0.04, 0.06, 0.09, 0.18, 0.13]
who really loves my	my	0.34	family	0.17	[0.03, 0.34, 0.03, 0.03, 0.03, 0.06, 0.09, 0.22, 0.17]

Sentence = "hi my name is harshu who really loves my family"

• Total words (tokens) = 10
• Context Length ($T$) = 4
• Batch Size ($B$) = 2
• Training Sets ($N$) = 24
• Batch Size Number of context-length input sequences passed to the model in one batch.
  • eg batch 1 has 2 (hi my name is & my name is harshu)
• context length is length of 1 input eg (my name is harshu) =4 tokens =4
• no. of training sets that can be created =

$$\text{Total Training Sets} = (\text{Total Tokens} - \text{Context Length}) \times \text{Context Length}$$ $$\text{Number of Batches} = \frac{\text{Number of Training Sets}} {\text{Context Length} \times \text{Batch Size}} = \frac{24}{4 \times 2} = 3$$

If we train for $E$ epochs:

$$\text{Total Batch Iterations} = \text{Number of Batches} \times \text{Epochs}$$

Example ($Epochs=5$):

$$\text{Total Batch Iterations} = 3 \times 5 = 15$$