Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76957.46132437699,
-      "required_pages_per_second_upper_estimate": 0.005324095740332254
+      "required_token_positions_per_second": 77272.8299493907,
+      "required_pages_per_second_upper_estimate": 0.005345913673566729
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9619.682665547123,
-      "required_pages_per_second_upper_estimate": 0.004027560250247491
+      "required_token_positions_per_second": 9659.103743673837,
+      "required_pages_per_second_upper_estimate": 0.004044065032453425
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19239.365331094246,
-      "required_pages_per_second_upper_estimate": 0.00400705242228358
+      "required_token_positions_per_second": 19318.207487347674,
+      "required_pages_per_second_upper_estimate": 0.004023473164223688
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28859.047996641373,
-      "required_pages_per_second_upper_estimate": 0.0017300558621838181
+      "required_token_positions_per_second": 28977.31123102151,
+      "required_pages_per_second_upper_estimate": 0.0017371455625074042
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28859.047996641373,
-      "required_pages_per_second_upper_estimate": 0.002243664044201902
+      "required_token_positions_per_second": 28977.31123102151,
+      "required_pages_per_second_upper_estimate": 0.0022528584904899646
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28859.047996641373,
-      "required_pages_per_second_upper_estimate": 0.37017288126937026
+      "required_token_positions_per_second": 28977.31123102151,
+      "required_pages_per_second_upper_estimate": 0.37168983505883085
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15815325870312336,
-    "producer_p95_seconds": 0.3763871119954274,
-    "gpu_update_seconds": 0.9580357606812839,
-    "producer_batches_per_second": 6.322980684686019,
-    "gpu_batches_per_second": 1.0438023725637071,
-    "producer_headroom": 6.057641610026236,
+    "producer_batch_seconds": 0.3127437677421767,
+    "producer_p95_seconds": 0.37166344500292325,
+    "gpu_update_seconds": 0.9541257910223767,
+    "producer_batches_per_second": 3.197505763965827,
+    "gpu_batches_per_second": 1.0480798332979424,
+    "producer_headroom": 3.0508227163425037,
     "sustainable": true,
-    "buffer_seconds": 60.35625292292089,
+    "buffer_seconds": 60.10992483440973,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 466853.83060173556,
-    "upload_bytes_per_second": 5467822.850568318,
+    "generation_bytes_per_second": 466602.93139571714,
+    "upload_bytes_per_second": 5485036.791124654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.17596424616862,
+    "estimated_full_shards_per_hour": 100.12212711719165,
     "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": 1.2989959650003584,
-      "bytes": 11854087,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0731057049997617,
       "bytes": 5340693,
@@ -562,6 +556,12 @@
       "bytes": 1224559,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.1411752389976755,
+      "bytes": 12344879,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -422,3 +422,4 @@
 {"step": 599300, "loss": 0.573486328125, "lr": 0.0003, "grad_norm": 0.18624040246734963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87719923972, "tokens_per_second": 153211.06162361678, "data_wait_s": 1.0664000001270324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599400, "loss": 0.582827627658844, "lr": 0.0003, "grad_norm": 0.1467533053899549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87734562111, "tokens_per_second": 151372.81203418892, "data_wait_s": 1.3983997632749379e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 599500, "loss": 0.5904302597045898, "lr": 0.0003, "grad_norm": 0.1590978193353224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87749038407, "tokens_per_second": 150917.65901102405, "data_wait_s": 1.0568997822701931e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 599600, "loss": 0.6239014267921448, "lr": 0.0003, "grad_norm": 0.1821307780857641, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87763667203, "tokens_per_second": 151769.4706270516, "data_wait_s": 1.1385003745090216e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8273f8d1ded0a579eea72bea03c35eed37c43f24e71efc66535f0de8acfc5c20
-size 904461
+oid sha256:5d1593f7c806268a33a5fa126cfc92a817045401b364efb6888417f33d6e9645
+size 906579