[GH-ISSUE #776] figure pass + ml ops feedback #4196

New Issue

GiteaMirror · 2026-04-19T12:12:19-05:00

GiteaMirror commented

2026-04-19 12:12:19 -05:00

Originally created by @18jeffreyma on GitHub (Mar 25, 2025).
Original GitHub issue: https://github.com/harvard-edge/cs249r_book/issues/776

Originally assigned to: @18jeffreyma on GitHub.

2.5.2 could use some references or images to break the flow
2.6.1 (Hybrid ML) could use a diagram for some of the hybrid techniques
2.2.3 could use a figure on the challenges of cloudML
2.7.3 (systems consideration layer): feels like a bunch of text: Add bolding for each section?
2.8: Table 2.2 doesn’t render well on PDF, Figure also doesn’t render well on PDF
3.3.6: Figure 3.9 caption is wrong (EPOCHS)
3.4.1: Perceptron In Figure 3.11 should come much further forward in PDF.
3.4.3.2 Design tradeoffs could use a diagram (i.e. wide versus deep networks)
3.5.1 should also use a diagram to show forward backwards prop etc. Maybe unsure on this, maybe place this at 3.5.4 or 3.5.5.3 and show a diagram showing autograd.
3.6.1 inference fundamentals: https://images.contentstack.io/v3/assets/blt71da4c740e00faaa/blt8226bbb9a1fbbfdf/5eea5e69146cc4629e1acebd/Training-vs-Inference-Graphic.jpg
3.7 Case Study https://production-media.paperswithcode.com/datasets/USPS-0000001055-6cd416b0_D96Rryg.jpg usps digit examples for 3.7
4.3: put an initial cnn diagram: https://miro.medium.com/v2/resize:fit:1400/1*7_BCJFzekmPXmJQVRdDgwg.png
4.5: https://jalammar.github.io/images/t/transformer_self-attention_visualization.png
4.6.4 for skip connections, show this figure https://production-media.paperswithcode.com/methods/resnet-e1548261477164.png
4.7.1 https://www.researchgate.net/profile/Wentai-Zhang-6/publication/342112418/figure/fig1/AS:902354379481088@1592149457294/The-im2col-GEMM-explicit-GEMM-method.png for im2col
4.7.3 add example all gather diagram? https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/_images/allgather.png
Bottom of 4.8 seems broken in PDF.
5.4.2/3 needs a diagram i think, unsure what it should be. Alternatively, move Figure 5.5 up.
5.5 needs a flow diagram showing how model development, training eval, etc.
5.6 needs a tinyml deployment diagram as an example?
5.7 needs a diagram of monitoring and maintenance or example reference.
5.8.1: Bold each of the roles to make it easier to read in PDF.
6.4.3: Maybe show amazon mechanical turk or a quarch screenshot?
6.5.2: https://d2908q01vomqb2.cloudfront.net/b6692ea5df920cad691c20319a6fffd7a4a766b8/2019/12/12/ETLandELTRedshift1.png for ETL vs ELTT
6.6.5: https://www.tensorflow.org/static/tfx/guide/images/prog_transform.png
6.8.5: https://www.tecton.ai/wp-content/uploads/2020/10/feature-store-tecton-blog-preview.png
6.9; https://drive.starcio.com/wp-content/uploads/2018/07/StarCIO-Data-Governance.jpg
7.3.3.1 for device placement: https://www.microsoft.com/en-us/research/wp-content/uploads/2020/09/Blog_DeepSpeed3_Figure2_highres.png
7.3.5: https://cdn-uploads.huggingface.co/production/uploads/5e73316106936008a9ee6523/D5FCD4NwDcr3m1rhz0mJ9.png for model and tensor parallelism
7.5.4: Orchestration https://www.altexsoft.com/static/blog-post/2023/11/fc3dd59e-998e-4288-aac1-f172a9204832.jpg
Weird figures at the end of 7.9 in book PDF?
8.5.2.4: https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQ0LsOl3extdLyE928tnoVzrF3Sa3hGBLGUtQ&s
8.5.2.6 https://docscontent.nvidia.com/dims4/default/f6210b6/2147483647/strip/true/crop/707x421+0+0/resize/707x421!/quality/90/?url=https%3A%2F%2Fk3-prod-nvidia-docs.s3.us-west-2.amazonaws.com%2Fbrightspot%2Fdita%2F00000189-949d-d46e-abe9-bcdf9f8c0000%2Fdeeplearning%2Fperformance%2Fmixed-precision-training%2Fgraphics%2Ftraining-iteration.png mixed precision divergence
8.5.3.4 https://iq.opengenus.org/content/images/2023/09/gradient.png
8.6.5.2: 3D-parallelism diagram from above, https://arxiv.org/abs/2502.12340
9.5.1: Maybe efficiency diagram for edge versus AV https://images.ctfassets.net/wcxs9ap8i19s/1Q0dAAdcsTRP4cdEEAqvSN/8b7686d3522f4bf59648b2fc7cebe5b0/why-latency-matters.png
9.7 needs diagrams, but unsure what would be best.
10.11 figure, move this up closer to 10.4.3.1
10.4.3.2: LoRA figure of determining appropriate rank
10.4.4: https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcRdBzXtU8UGC5wsB8HYAppgxmJ_-fNEmFksmA&s or https://theaisummer.com/static/16886a5459bdc62b845d82350e76b4ce/f058b/nat.png
Weird figures at the end of section 10 as well? Huge figures for some reason?
10.7.1 https://learn.microsoft.com/en-us/dotnet/machine-learning/media/ml-automl-workflow.png
Figure 11.4 maybe move up further?
Table 11.5 capitalise CUDA
11.5.1.1 tiling https://docscontent.nvidia.com/dita/00000189-949d-d46e-abe9-bcdf9f8c0000/deeplearning/performance/dl-performance-matrix-multiplication/graphics/tiled-outer-prod.svg
11.5.2.3 memory allocation? https://pytorch.org/assets/images/understanding-gpu-memory-1/fig6.png
11.6.1: https://user-images.githubusercontent.com/38183241/135726581-ee305818-c78a-439f-90b4-30cd1edbc1fe.png
11.6.1.2 row vs column major https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQJc9Zx5pJyQCwrSOPPFRGVoiFSIHgiLapE1A&s, https://craftofcoding.wordpress.com/wp-content/uploads/2017/02/rowcolumnarrays.jpg
11.6.2.2: Flash attention https://miro.medium.com/v2/resize:fit:2000/format:webp/1*i4tDdwgvGtXuTIyJpFUn8A.png
11.6.2.3: Blocked matmul https://www.cs.sfu.ca/~ashriram/Courses/CS7ARCH/hw/hw4/gemm_tiled.png
11.7.3.1: Diagram for showing graph optimization? https://danny270degree.blogspot.com/2018/06/xla-take-glance-to-see-graph-changes-in.html
BIG ISSUE COPYING FROM PDF DOES NOT WORK
11.7.6.1 overlapping: https://arxiv.org/html/2409.15241v1/x2.png
11.9 seems to need more diagrams, let’s brainstorm.
11.11 resources WIP
12.6 maybe some figures here from MLPerf and sprinkle them in?
12.8.2 is there some kind of ml perf power plot to put here?

Originally created by @18jeffreyma on GitHub (Mar 25, 2025). Original GitHub issue: https://github.com/harvard-edge/cs249r_book/issues/776 Originally assigned to: @18jeffreyma on GitHub. - [x] 2.5.2 could use some references or images to break the flow - [ ] 2.6.1 (Hybrid ML) could use a diagram for some of the hybrid techniques - [ ] 2.2.3 could use a figure on the challenges of cloudML - [x] 2.7.3 (systems consideration layer): feels like a bunch of text: Add bolding for each section? - [ ] 2.8: Table 2.2 doesn’t render well on PDF, Figure also doesn’t render well on PDF - [x] 3.3.6: Figure 3.9 caption is wrong (EPOCHS) - [x] 3.4.1: Perceptron In Figure 3.11 should come much further forward in PDF. - [x] 3.4.3.2 Design tradeoffs could use a diagram (i.e. wide versus deep networks) - [x] 3.5.1 should also use a diagram to show forward backwards prop etc. Maybe unsure on this, maybe place this at 3.5.4 or 3.5.5.3 and show a diagram showing autograd. - [x] 3.6.1 inference fundamentals: [https://images.contentstack.io/v3/assets/blt71da4c740e00faaa/blt8226bbb9a1fbbfdf/5eea5e69146cc4629e1acebd/Training-vs-Inference-Graphic.jpg](https://images.contentstack.io/v3/assets/blt71da4c740e00faaa/blt8226bbb9a1fbbfdf/5eea5e69146cc4629e1acebd/Training-vs-Inference-Graphic.jpg) - [x] 3.7 Case Study [https://production-media.paperswithcode.com/datasets/USPS-0000001055-6cd416b0\_D96Rryg.jpg](https://production-media.paperswithcode.com/datasets/USPS-0000001055-6cd416b0_D96Rryg.jpg) usps digit examples for 3.7 - [x] 4.3: put an initial cnn diagram: [https://miro.medium.com/v2/resize:fit:1400/1\*7\_BCJFzekmPXmJQVRdDgwg.png](https://miro.medium.com/v2/resize:fit:1400/1*7_BCJFzekmPXmJQVRdDgwg.png) - [x] 4.5: [https://jalammar.github.io/images/t/transformer\_self-attention\_visualization.png](https://jalammar.github.io/images/t/transformer_self-attention_visualization.png) - [x] 4.6.4 for skip connections, show this figure [https://production-media.paperswithcode.com/methods/resnet-e1548261477164.png](https://production-media.paperswithcode.com/methods/resnet-e1548261477164.png) - [x] 4.7.1 [https://www.researchgate.net/profile/Wentai-Zhang-6/publication/342112418/figure/fig1/AS:902354379481088@1592149457294/The-im2col-GEMM-explicit-GEMM-method.png](https://www.researchgate.net/profile/Wentai-Zhang-6/publication/342112418/figure/fig1/AS:902354379481088@1592149457294/The-im2col-GEMM-explicit-GEMM-method.png) for im2col - [x] 4.7.3 add example all gather diagram? [https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/\_images/allgather.png](https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/_images/allgather.png) - [ ] Bottom of 4.8 seems broken in PDF. - [x] 5.4.2/3 needs a diagram i think, unsure what it should be. Alternatively, move Figure 5.5 up. - [ ] 5.5 needs a flow diagram showing how model development, training eval, etc. - [ ] 5.6 needs a tinyml deployment diagram as an example? - [ ] 5.7 needs a diagram of monitoring and maintenance or example reference. - [ ] 5.8.1: Bold each of the roles to make it easier to read in PDF. - [ ] 6.4.3: Maybe show amazon mechanical turk or a quarch screenshot? - [x] 6.5.2: [https://d2908q01vomqb2.cloudfront.net/b6692ea5df920cad691c20319a6fffd7a4a766b8/2019/12/12/ETLandELTRedshift1.png](https://d2908q01vomqb2.cloudfront.net/b6692ea5df920cad691c20319a6fffd7a4a766b8/2019/12/12/ETLandELTRedshift1.png) for ETL vs ELTT - [x] 6.6.5: [https://www.tensorflow.org/static/tfx/guide/images/prog\_transform.png](https://www.tensorflow.org/static/tfx/guide/images/prog_transform.png) - [x] 6.8.5: [https://www.tecton.ai/wp-content/uploads/2020/10/feature-store-tecton-blog-preview.png](https://www.tecton.ai/wp-content/uploads/2020/10/feature-store-tecton-blog-preview.png) - [x] 6.9; [https://drive.starcio.com/wp-content/uploads/2018/07/StarCIO-Data-Governance.jpg](https://drive.starcio.com/wp-content/uploads/2018/07/StarCIO-Data-Governance.jpg) - [x] 7.3.3.1 for device placement: [https://www.microsoft.com/en-us/research/wp-content/uploads/2020/09/Blog\_DeepSpeed3\_Figure2\_highres.png](https://www.microsoft.com/en-us/research/wp-content/uploads/2020/09/Blog_DeepSpeed3_Figure2_highres.png) - [x] 7.3.5: [https://cdn-uploads.huggingface.co/production/uploads/5e73316106936008a9ee6523/D5FCD4NwDcr3m1rhz0mJ9.png](https://cdn-uploads.huggingface.co/production/uploads/5e73316106936008a9ee6523/D5FCD4NwDcr3m1rhz0mJ9.png) for model and tensor parallelism - [x] 7.5.4: Orchestration [https://www.altexsoft.com/static/blog-post/2023/11/fc3dd59e-998e-4288-aac1-f172a9204832.jpg](https://www.altexsoft.com/static/blog-post/2023/11/fc3dd59e-998e-4288-aac1-f172a9204832.jpg) - [x] Weird figures at the end of 7.9 in book PDF? - [x] 8.5.2.4: [https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQ0LsOl3extdLyE928tnoVzrF3Sa3hGBLGUtQ\&s](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQ0LsOl3extdLyE928tnoVzrF3Sa3hGBLGUtQ&s) - [x] 8.5.2.6 [https://docscontent.nvidia.com/dims4/default/f6210b6/2147483647/strip/true/crop/707x421+0+0/resize/707x421\!/quality/90/?url=https%3A%2F%2Fk3-prod-nvidia-docs.s3.us-west-2.amazonaws.com%2Fbrightspot%2Fdita%2F00000189-949d-d46e-abe9-bcdf9f8c0000%2Fdeeplearning%2Fperformance%2Fmixed-precision-training%2Fgraphics%2Ftraining-iteration.png](https://docscontent.nvidia.com/dims4/default/f6210b6/2147483647/strip/true/crop/707x421+0+0/resize/707x421!/quality/90/?url=https%3A%2F%2Fk3-prod-nvidia-docs.s3.us-west-2.amazonaws.com%2Fbrightspot%2Fdita%2F00000189-949d-d46e-abe9-bcdf9f8c0000%2Fdeeplearning%2Fperformance%2Fmixed-precision-training%2Fgraphics%2Ftraining-iteration.png) mixed precision divergence - [x] 8.5.3.4 [https://iq.opengenus.org/content/images/2023/09/gradient.png](https://iq.opengenus.org/content/images/2023/09/gradient.png) - [ ] 8.6.5.2: 3D-parallelism diagram from above, [https://arxiv.org/abs/2502.12340](https://arxiv.org/abs/2502.12340) - [x] 9.5.1: Maybe efficiency diagram for edge versus AV [https://images.ctfassets.net/wcxs9ap8i19s/1Q0dAAdcsTRP4cdEEAqvSN/8b7686d3522f4bf59648b2fc7cebe5b0/why-latency-matters.png](https://images.ctfassets.net/wcxs9ap8i19s/1Q0dAAdcsTRP4cdEEAqvSN/8b7686d3522f4bf59648b2fc7cebe5b0/why-latency-matters.png) - [ ] 9.7 needs diagrams, but unsure what would be best. - [x] 10.11 figure, move this up closer to 10.4.3.1 - [x] 10.4.3.2: LoRA figure of determining appropriate rank - [x] 10.4.4: [https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcRdBzXtU8UGC5wsB8HYAppgxmJ\_-fNEmFksmA\&s](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcRdBzXtU8UGC5wsB8HYAppgxmJ_-fNEmFksmA&s) or [https://theaisummer.com/static/16886a5459bdc62b845d82350e76b4ce/f058b/nat.png](https://theaisummer.com/static/16886a5459bdc62b845d82350e76b4ce/f058b/nat.png) - [x] Weird figures at the end of section 10 as well? Huge figures for some reason? - [x] 10.7.1 [https://learn.microsoft.com/en-us/dotnet/machine-learning/media/ml-automl-workflow.png](https://learn.microsoft.com/en-us/dotnet/machine-learning/media/ml-automl-workflow.png) - [ ] Figure 11.4 maybe move up further? - [x] Table 11.5 capitalise CUDA - [x] 11.5.1.1 tiling [https://docscontent.nvidia.com/dita/00000189-949d-d46e-abe9-bcdf9f8c0000/deeplearning/performance/dl-performance-matrix-multiplication/graphics/tiled-outer-prod.svg](https://docscontent.nvidia.com/dita/00000189-949d-d46e-abe9-bcdf9f8c0000/deeplearning/performance/dl-performance-matrix-multiplication/graphics/tiled-outer-prod.svg) - [ ] 11.5.2.3 memory allocation? [https://pytorch.org/assets/images/understanding-gpu-memory-1/fig6.png](https://pytorch.org/assets/images/understanding-gpu-memory-1/fig6.png) - [x] 11.6.1: [https://user-images.githubusercontent.com/38183241/135726581-ee305818-c78a-439f-90b4-30cd1edbc1fe.png](https://user-images.githubusercontent.com/38183241/135726581-ee305818-c78a-439f-90b4-30cd1edbc1fe.png) - [x] 11.6.1.2 row vs column major [https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQJc9Zx5pJyQCwrSOPPFRGVoiFSIHgiLapE1A\&s](https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQJc9Zx5pJyQCwrSOPPFRGVoiFSIHgiLapE1A&s), [https://craftofcoding.wordpress.com/wp-content/uploads/2017/02/rowcolumnarrays.jpg](https://craftofcoding.wordpress.com/wp-content/uploads/2017/02/rowcolumnarrays.jpg) - [x] 11.6.2.2: Flash attention [https://miro.medium.com/v2/resize:fit:2000/format:webp/1\*i4tDdwgvGtXuTIyJpFUn8A.png](https://miro.medium.com/v2/resize:fit:2000/format:webp/1*i4tDdwgvGtXuTIyJpFUn8A.png) - [x] 11.6.2.3: Blocked matmul [https://www.cs.sfu.ca/\~ashriram/Courses/CS7ARCH/hw/hw4/gemm\_tiled.png](https://www.cs.sfu.ca/~ashriram/Courses/CS7ARCH/hw/hw4/gemm_tiled.png) - [x] 11.7.3.1: Diagram for showing graph optimization? [https://danny270degree.blogspot.com/2018/06/xla-take-glance-to-see-graph-changes-in.html](https://danny270degree.blogspot.com/2018/06/xla-take-glance-to-see-graph-changes-in.html) - [x] BIG ISSUE COPYING FROM PDF DOES NOT WORK - [x] 11.7.6.1 overlapping: [https://arxiv.org/html/2409.15241v1/x2.png](https://arxiv.org/html/2409.15241v1/x2.png) - [ ] 11.9 seems to need more diagrams, let’s brainstorm. - [ ] 11.11 resources WIP - [x] 12.6 maybe some figures here from MLPerf and sprinkle them in? - [ ] 12.8.2 is there some kind of ml perf power plot to put here?

GiteaMirror added the area: book type: improvement labels 2026-04-19 12:12:19 -05:00

GiteaMirror closed this issue

2026-04-19 12:12:20 -05:00

GiteaMirror commented

2026-04-19 12:12:22 -05:00

@18jeffreyma commented on GitHub (Mar 25, 2025):

Chapter 13 feedback;

Diagram 13.2.2 https://ml-ops.org/img/mlops-loop-en.jpg
Figure 13.3 needs higher resolution on the textbook PDF
13.3.4: Discussion on predefining benchmarks and tasks of interest, and verifying that those tasks are correlated with desired real world performance. https://www.databricks.com/blog/calibrating-mosaic-evaluation-gauntlet
Diagram for data drift: https://cdn.prod.website-files.com/660ef16a9e0687d9cc27474a/662c3c83dc614ac9ad2502f6_653fd39e0ee7f43f1055acc9_model_monitoring4.png
Data engineer/scientist figure? https://miro.medium.com/v2/resize:fit:1400/1*iW95hu86bn_FXskUWh3f7A.jpeg https://cdn.prod.website-files.com/63119622d2a6edf1d171e0bc/6729dd213c445f8632403242_6729dcf95f301e553fc6110c_Roles%2520and%2520Responsibilities%2520of%2520Data%2520Scientist.webp
Alternatively a big figure showing how product managers + datascientists + ml engineers interact.
13.6.3 diagram: https://mlops-guide.github.io/MLOps/Monitoring/monitoring_iceberg.png
13.7 oura ring: https://media-cldnry.s-nbcnews.com/image/upload/t_fit-1240w,f_auto,q_auto:best/rockcms/2024-07/240612-Oura-ring-ode-bd-main-10a116.jpg

@18jeffreyma commented on GitHub (Mar 25, 2025): Chapter 13 feedback; - [x] Diagram 13.2.2 [https://ml-ops.org/img/mlops-loop-en.jpg](https://ml-ops.org/img/mlops-loop-en.jpg) - [x] Figure 13.3 needs higher resolution on the textbook PDF - [ ] 13.3.4: Discussion on predefining benchmarks and tasks of interest, and verifying that those tasks are correlated with desired real world performance. [https://www.databricks.com/blog/calibrating-mosaic-evaluation-gauntlet](https://www.databricks.com/blog/calibrating-mosaic-evaluation-gauntlet) - [x] Diagram for data drift: [https://cdn.prod.website-files.com/660ef16a9e0687d9cc27474a/662c3c83dc614ac9ad2502f6\_653fd39e0ee7f43f1055acc9\_model\_monitoring4.png](https://cdn.prod.website-files.com/660ef16a9e0687d9cc27474a/662c3c83dc614ac9ad2502f6_653fd39e0ee7f43f1055acc9_model_monitoring4.png) - [x] Data engineer/scientist figure? [https://miro.medium.com/v2/resize:fit:1400/1\*iW95hu86bn\_FXskUWh3f7A.jpeg](https://miro.medium.com/v2/resize:fit:1400/1*iW95hu86bn_FXskUWh3f7A.jpeg) [https://cdn.prod.website-files.com/63119622d2a6edf1d171e0bc/6729dd213c445f8632403242\_6729dcf95f301e553fc6110c\_Roles%2520and%2520Responsibilities%2520of%2520Data%2520Scientist.webp](https://cdn.prod.website-files.com/63119622d2a6edf1d171e0bc/6729dd213c445f8632403242_6729dcf95f301e553fc6110c_Roles%2520and%2520Responsibilities%2520of%2520Data%2520Scientist.webp) - [x] Alternatively a big figure showing how product managers \+ datascientists \+ ml engineers interact. - [x] 13.6.3 diagram: [https://mlops-guide.github.io/MLOps/Monitoring/monitoring\_iceberg.png](https://mlops-guide.github.io/MLOps/Monitoring/monitoring_iceberg.png) - [x] 13.7 oura ring: [https://media-cldnry.s-nbcnews.com/image/upload/t\_fit-1240w,f\_auto,q\_auto:best/rockcms/2024-07/240612-Oura-ring-ode-bd-main-10a116.jpg](https://media-cldnry.s-nbcnews.com/image/upload/t_fit-1240w,f_auto,q_auto:best/rockcms/2024-07/240612-Oura-ring-ode-bd-main-10a116.jpg)

GiteaMirror commented

2026-04-19 12:12:22 -05:00

@18jeffreyma commented on GitHub (May 29, 2025):

ive copied over the important ones for PDF rendering to https://github.com/harvard-edge/cs249r_book/issues/833, everything else here is "nice to have" but requires more long term thinking

@18jeffreyma commented on GitHub (May 29, 2025): ive copied over the important ones for PDF rendering to https://github.com/harvard-edge/cs249r_book/issues/833, everything else here is "nice to have" but requires more long term thinking

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/cs249r_book#4196