Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76069.06281119368,
-      "required_pages_per_second_upper_estimate": 0.006593473569197194
+      "required_token_positions_per_second": 76127.35069493848,
+      "required_pages_per_second_upper_estimate": 0.006598525815230901
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9508.63285139921,
-      "required_pages_per_second_upper_estimate": 0.003793135811153347
+      "required_token_positions_per_second": 9515.91883686731,
+      "required_pages_per_second_upper_estimate": 0.003796042299691762
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19017.26570279842,
-      "required_pages_per_second_upper_estimate": 0.004399309727897212
+      "required_token_positions_per_second": 19031.83767373462,
+      "required_pages_per_second_upper_estimate": 0.004402680696915372
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28525.898554197633,
-      "required_pages_per_second_upper_estimate": 0.0017370775581275516
+      "required_token_positions_per_second": 28547.756510601925,
+      "required_pages_per_second_upper_estimate": 0.0017384085929927388
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28525.898554197633,
-      "required_pages_per_second_upper_estimate": 0.0021960026354433003
+      "required_token_positions_per_second": 28547.756510601925,
+      "required_pages_per_second_upper_estimate": 0.0021976853214339983
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28525.898554197633,
-      "required_pages_per_second_upper_estimate": 0.3541961900020814
+      "required_token_positions_per_second": 28547.756510601925,
+      "required_pages_per_second_upper_estimate": 0.3544675926667485
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10617879225785032,
-    "producer_p95_seconds": 0.36039348499980406,
-    "gpu_update_seconds": 0.9692245082997237,
-    "producer_batches_per_second": 9.418076611490795,
-    "gpu_batches_per_second": 1.0317526965493935,
-    "producer_headroom": 9.128230672901294,
+    "producer_batch_seconds": 0.12316699090627026,
+    "producer_p95_seconds": 0.3602621399986674,
+    "gpu_update_seconds": 0.9684824091074274,
+    "producer_batches_per_second": 8.11905846397593,
+    "gpu_batches_per_second": 1.0325432765698035,
+    "producer_headroom": 7.863165300875458,
     "sustainable": true,
-    "buffer_seconds": 60.09191951458287,
+    "buffer_seconds": 60.0459093646605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 546625.9737170503,
-    "upload_bytes_per_second": 5222114.950742523,
+    "generation_bytes_per_second": 543837.8009776565,
+    "upload_bytes_per_second": 5132226.018885838,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.29320915826446,
+    "estimated_full_shards_per_hour": 116.69493219373008,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.3402203920013562,
-      "bytes": 12178587,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 0.8894950190006057,
       "bytes": 1180133,
@@ -562,6 +556,12 @@
       "bytes": 6247577,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.892657821001194,
+      "bytes": 2615407,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 146800, "loss": 0.5871815280988812, "lr": 0.0003, "grad_norm": 0.1341965422469446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21483789927, "tokens_per_second": 151155.09506963321, "data_wait_s": 9.743998816702515e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 146900, "loss": 0.6355623006820679, "lr": 0.0003, "grad_norm": 0.1646902456067799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21498590125, "tokens_per_second": 150710.4243415569, "data_wait_s": 1.0867999662877992e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147000, "loss": 0.5945109017193317, "lr": 0.0003, "grad_norm": 0.13763741233161803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21513190824, "tokens_per_second": 149468.14629271862, "data_wait_s": 9.603001672076061e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 147100, "loss": 0.5724699255079031, "lr": 0.0003, "grad_norm": 0.1468198003576462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21527672095, "tokens_per_second": 147824.88346666755, "data_wait_s": 1.1464999261079356e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:106f9b5e489f47045b5b13c65df6c1827c9bdacb06a1c8d91d7fe7ea7676f6e8
-size 682623
+oid sha256:7a6170d8182f8215502fe6a3845b239227ab9c191a133e03325ece410447ffee
+size 684741