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": 77384.08492129024,
-      "required_pages_per_second_upper_estimate": 0.0053536105506698396
+      "required_token_positions_per_second": 77186.01818108233,
+      "required_pages_per_second_upper_estimate": 0.0053399078340041015
     },
     {
       "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": 9673.01061516128,
-      "required_pages_per_second_upper_estimate": 0.004049887549136717
+      "required_token_positions_per_second": 9648.25227263529,
+      "required_pages_per_second_upper_estimate": 0.004039521748134069
     },
     {
       "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": 19346.02123032256,
-      "required_pages_per_second_upper_estimate": 0.004029266033387629
+      "required_token_positions_per_second": 19296.50454527058,
+      "required_pages_per_second_upper_estimate": 0.004018953013734101
     },
     {
       "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": 29019.03184548384,
-      "required_pages_per_second_upper_estimate": 0.0017396466496407313
+      "required_token_positions_per_second": 28944.756817905873,
+      "required_pages_per_second_upper_estimate": 0.0017351939751488325
     },
     {
       "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": 29019.03184548384,
-      "required_pages_per_second_upper_estimate": 0.0022561020847548215
+      "required_token_positions_per_second": 28944.756817905873,
+      "required_pages_per_second_upper_estimate": 0.002250327528061954
     },
     {
       "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": 29019.03184548384,
-      "required_pages_per_second_upper_estimate": 0.3722249823050479
+      "required_token_positions_per_second": 28944.756817905873,
+      "required_pages_per_second_upper_estimate": 0.3712722620015889
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4221679549064561,
-    "producer_p95_seconds": 0.35173528399900533,
-    "gpu_update_seconds": 0.9527540459383999,
-    "producer_batches_per_second": 2.368725499834728,
-    "gpu_batches_per_second": 1.0495888254298265,
-    "producer_headroom": 2.2568128036849955,
+    "producer_batch_seconds": 0.5511397114613601,
+    "producer_p95_seconds": 0.3740556490010931,
+    "gpu_update_seconds": 0.9551989043796812,
+    "producer_batches_per_second": 1.8144219681584477,
+    "gpu_batches_per_second": 1.0469023733328224,
+    "producer_headroom": 1.7331338760673742,
     "sustainable": true,
-    "buffer_seconds": 60.02350489411919,
+    "buffer_seconds": 60.17753097591992,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301563.0620923001,
-    "upload_bytes_per_second": 4752808.928778377,
+    "generation_bytes_per_second": 272557.31630401284,
+    "upload_bytes_per_second": 4791883.095739071,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.70841309620621,
+    "estimated_full_shards_per_hour": 58.48445526924409,
     "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": 1.300822404999053,
-      "bytes": 6046016,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.4766436999998405,
       "bytes": 10238885,
@@ -562,6 +556,12 @@
       "bytes": 11898828,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.6147537899960298,
+      "bytes": 11121609,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -432,3 +432,4 @@
 {"step": 688200, "loss": 0.5151166915893555, "lr": 0.0001, "grad_norm": 0.2120875927685729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100733890275, "tokens_per_second": 153225.4642736849, "data_wait_s": 1.7035999917425215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688300, "loss": 0.5191828608512878, "lr": 0.0001, "grad_norm": 0.1867570862786017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100748617566, "tokens_per_second": 153053.5687605134, "data_wait_s": 1.6682999557815492e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688400, "loss": 0.5875371098518372, "lr": 0.0001, "grad_norm": 0.213001294053636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100763397324, "tokens_per_second": 141106.5089613713, "data_wait_s": 1.1366995749995112e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688500, "loss": 0.5560246706008911, "lr": 0.0001, "grad_norm": 0.18670503836679223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100778019023, "tokens_per_second": 152148.12010560403, "data_wait_s": 1.3844997738488019e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9b67aa33694d0a11c13d957a201033b5ef4f12c23992a0291676b86841a1571
-size 925641
+oid sha256:4d2f4b9c1aa9997a0276135d34cb28ad4bb92925e6e564398909b8684a8fc3c5
+size 927759