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": 75504.79548572049,
-      "required_pages_per_second_upper_estimate": 0.0065445643075476385
+      "required_token_positions_per_second": 75590.71132067639,
+      "required_pages_per_second_upper_estimate": 0.006552011274369923
     },
     {
       "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": 9438.099435715061,
-      "required_pages_per_second_upper_estimate": 0.0037649989770683983
+      "required_token_positions_per_second": 9448.838915084549,
+      "required_pages_per_second_upper_estimate": 0.0037692831159584126
     },
     {
       "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": 18876.198871430122,
-      "required_pages_per_second_upper_estimate": 0.004366676399151594
+      "required_token_positions_per_second": 18897.677830169097,
+      "required_pages_per_second_upper_estimate": 0.004371645178239092
     },
     {
       "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": 28314.298307145185,
-      "required_pages_per_second_upper_estimate": 0.0017241922132627547
+      "required_token_positions_per_second": 28346.516745253644,
+      "required_pages_per_second_upper_estimate": 0.0017261541471064788
     },
     {
       "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": 28314.298307145185,
-      "required_pages_per_second_upper_estimate": 0.0021797130626782296
+      "required_token_positions_per_second": 28346.516745253644,
+      "required_pages_per_second_upper_estimate": 0.0021821933272301633
     },
     {
       "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": 28314.298307145185,
-      "required_pages_per_second_upper_estimate": 0.351568823114161
+      "required_token_positions_per_second": 28346.516745253644,
+      "required_pages_per_second_upper_estimate": 0.3519688682872921
     }
   ],
   "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": 0.11515667031248711,
-    "producer_p95_seconds": 0.2873591250008758,
-    "gpu_update_seconds": 0.976467779638493,
-    "producer_batches_per_second": 8.683821764613528,
-    "gpu_batches_per_second": 1.0240993311322768,
-    "producer_headroom": 8.47947215726859,
+    "producer_batch_seconds": 0.1002805079295257,
+    "producer_p95_seconds": 0.24397912999847904,
+    "gpu_update_seconds": 0.975357933691426,
+    "producer_batches_per_second": 9.972027671646535,
+    "gpu_batches_per_second": 1.0252646392235838,
+    "producer_headroom": 9.726296304530885,
     "sustainable": true,
-    "buffer_seconds": 60.54100233758656,
+    "buffer_seconds": 60.47219188886841,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375493.1483818882,
-    "upload_bytes_per_second": 5938363.140113106,
+    "generation_bytes_per_second": 385525.12457990967,
+    "upload_bytes_per_second": 5827243.82067359,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.57208860962376,
+    "estimated_full_shards_per_hour": 82.72471716926543,
     "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": 1.2269071499995334,
-      "bytes": 11690462,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.9768262770003275,
       "bytes": 1050575,
@@ -562,6 +556,12 @@
       "bytes": 6343213,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.028006202999677,
+      "bytes": 2590442,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -130,3 +130,4 @@
 {"step": 128000, "loss": 0.6035799086093903, "lr": 0.0003, "grad_norm": 0.13391310087245253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18732618299, "tokens_per_second": 150480.36727232047, "data_wait_s": 1.367599907098338e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 128100, "loss": 0.6263636723160744, "lr": 0.0003, "grad_norm": 0.13098580312341396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18747193942, "tokens_per_second": 148760.9482322472, "data_wait_s": 1.6726000467315316e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 128200, "loss": 0.5950319953262806, "lr": 0.0003, "grad_norm": 0.1340751081664851, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18762040195, "tokens_per_second": 150997.25168156862, "data_wait_s": 1.2567999874590896e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 128300, "loss": 0.570756122469902, "lr": 0.0003, "grad_norm": 0.14187998051100334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18776559922, "tokens_per_second": 147947.98340237938, "data_wait_s": 8.572000297135673e-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 sha256:6a3d0168e11c0b98276c38e80364b3395a2b139f3ddb7f1d36484def7a734ee7
-size 281307
+oid sha256:e061a6828110cc98490ec1f75979a969c069730dcd6edf8429f6da863ce1ab60
+size 283425