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": 75738.16451328064,
-      "required_pages_per_second_upper_estimate": 0.00656479214338816
+      "required_token_positions_per_second": 75528.77367106377,
+      "required_pages_per_second_upper_estimate": 0.006546642675881039
     },
     {
       "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": 9467.27056416008,
-      "required_pages_per_second_upper_estimate": 0.003776635776352354
+      "required_token_positions_per_second": 9441.096708882971,
+      "required_pages_per_second_upper_estimate": 0.003766194634148305
     },
     {
       "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": 18934.54112832016,
-      "required_pages_per_second_upper_estimate": 0.004380172853494452
+      "required_token_positions_per_second": 18882.193417765942,
+      "required_pages_per_second_upper_estimate": 0.004368063131945973
     },
     {
       "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": 28401.811692480245,
-      "required_pages_per_second_upper_estimate": 0.0017295213192824116
+      "required_token_positions_per_second": 28323.290126648913,
+      "required_pages_per_second_upper_estimate": 0.0017247397678941006
     },
     {
       "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": 28401.811692480245,
-      "required_pages_per_second_upper_estimate": 0.002186450085333878
+      "required_token_positions_per_second": 28323.290126648913,
+      "required_pages_per_second_upper_estimate": 0.002180405277834576
     },
     {
       "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": 28401.811692480245,
-      "required_pages_per_second_upper_estimate": 0.3526554464715627
+      "required_token_positions_per_second": 28323.290126648913,
+      "required_pages_per_second_upper_estimate": 0.3516804714187133
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05400955532816454,
-    "producer_p95_seconds": 0.18607703399902675,
-    "gpu_update_seconds": 0.9734590278732175,
-    "producer_batches_per_second": 18.515242236747813,
-    "gpu_batches_per_second": 1.0272646011458422,
-    "producer_headroom": 18.023829708621665,
+    "producer_batch_seconds": 0.06922779871096907,
+    "producer_p95_seconds": 0.18515474100058782,
+    "gpu_update_seconds": 0.9761577795648273,
+    "producer_batches_per_second": 14.445064246157392,
+    "gpu_batches_per_second": 1.024424556085392,
+    "producer_headroom": 14.100661840200274,
     "sustainable": true,
-    "buffer_seconds": 60.35445972813948,
+    "buffer_seconds": 60.521782333019296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394878.46069678426,
-    "upload_bytes_per_second": 5997526.5266562095,
+    "generation_bytes_per_second": 475696.2186279059,
+    "upload_bytes_per_second": 5988381.0431731995,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.73172536542555,
+    "estimated_full_shards_per_hour": 102.0733348763264,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 6185442,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0889191619990015,
+      "bytes": 6163490,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -90,3 +90,4 @@
 {"step": 124000, "loss": 0.627118244767189, "lr": 0.0003, "grad_norm": 0.12970198441350902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18145808615, "tokens_per_second": 149335.03056101763, "data_wait_s": 1.2071999663021415e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124100, "loss": 0.6504697818309069, "lr": 0.0003, "grad_norm": 0.1574087697589261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18160506122, "tokens_per_second": 151009.73140468044, "data_wait_s": 1.4882998584653251e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124200, "loss": 0.6450304668396711, "lr": 0.0003, "grad_norm": 0.1337577137552498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18175258782, "tokens_per_second": 151434.2018835993, "data_wait_s": 1.2652000805246644e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 124300, "loss": 0.6086334884166718, "lr": 0.0003, "grad_norm": 0.1368946043999693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18190034351, "tokens_per_second": 151453.04840488223, "data_wait_s": 9.346998922410421e-06, "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 sha256:56ce6b92bc7ef8f11dd553b6f9a54601b80ce990e24a832ec44ceda08507699a
-size 195804
+oid sha256:eb2b20359ed0709f38f189526d880f2bb1208c51cd3331ed7c866eb12917b75c
+size 197922