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": 76955.31394533147,
-      "required_pages_per_second_upper_estimate": 0.005323947179667279
+      "required_token_positions_per_second": 76356.71216966122,
+      "required_pages_per_second_upper_estimate": 0.005282534519878932
     },
     {
       "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": 9619.414243166433,
-      "required_pages_per_second_upper_estimate": 0.004027447867401993
+      "required_token_positions_per_second": 9544.589021207652,
+      "required_pages_per_second_upper_estimate": 0.003996120109496166
     },
     {
       "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": 19238.828486332866,
-      "required_pages_per_second_upper_estimate": 0.0040069406116773326
+      "required_token_positions_per_second": 19089.178042415304,
+      "required_pages_per_second_upper_estimate": 0.003975772370756905
     },
     {
       "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": 28858.242729499296,
-      "required_pages_per_second_upper_estimate": 0.0017300075876482224
+      "required_token_positions_per_second": 28633.767063622956,
+      "required_pages_per_second_upper_estimate": 0.0017165506142334315
     },
     {
       "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": 28858.242729499296,
-      "required_pages_per_second_upper_estimate": 0.0022436014382236015
+      "required_token_positions_per_second": 28633.767063622956,
+      "required_pages_per_second_upper_estimate": 0.0022261494425658223
     },
     {
       "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": 28858.242729499296,
-      "required_pages_per_second_upper_estimate": 0.37016255216710015
+      "required_token_positions_per_second": 28633.767063622956,
+      "required_pages_per_second_upper_estimate": 0.36728321934843006
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08477259781227531,
-    "producer_p95_seconds": 0.2500173309999809,
-    "gpu_update_seconds": 0.9580624939348032,
-    "producer_batches_per_second": 11.796264663428744,
-    "gpu_batches_per_second": 1.0437732468713576,
-    "producer_headroom": 11.301558742559534,
+    "producer_batch_seconds": 0.48389465510942387,
+    "producer_p95_seconds": 0.2651760810003907,
+    "gpu_update_seconds": 0.965573266645893,
+    "producer_batches_per_second": 2.0665655002406846,
+    "gpu_batches_per_second": 1.0356541906692331,
+    "producer_headroom": 1.9954204008051017,
     "sustainable": true,
-    "buffer_seconds": 60.357937117892604,
+    "buffer_seconds": 60.831115798691265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330331.7014523559,
+    "generation_bytes_per_second": 301307.5052581362,
     "upload_bytes_per_second": 5060966.021529688,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.88149340322502,
+    "estimated_full_shards_per_hour": 64.65357654865326,
     "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": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -193,3 +193,4 @@
 {"step": 779300, "loss": 0.5392187833786011, "lr": 3e-05, "grad_norm": 0.2561384988112868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114067641327, "tokens_per_second": 153041.10515992413, "data_wait_s": 1.3837998267263174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779400, "loss": 0.5333099365234375, "lr": 3e-05, "grad_norm": 0.24759982815345763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114082340629, "tokens_per_second": 153000.9783801805, "data_wait_s": 1.1845000699395314e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779500, "loss": 0.5458994507789612, "lr": 3e-05, "grad_norm": 0.25431371027240035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114097047090, "tokens_per_second": 145218.2433116972, "data_wait_s": 1.3504002708941698e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 779600, "loss": 0.5156227946281433, "lr": 3e-05, "grad_norm": 0.26047371955239, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114111758568, "tokens_per_second": 151882.10459962077, "data_wait_s": 1.2558000889839604e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3c5d451738de547eb9651f08426fc348bec9cf00b74517ed6ba14404cc5e87f
-size 415524
+oid sha256:d18616cdaf0eaf3d0aca08b68222924b61ec16218984ef481c03c661167a0b56
+size 417642