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": 77324.43126543002,
-      "required_pages_per_second_upper_estimate": 0.005349483572341878
+      "required_token_positions_per_second": 77189.20408654222,
+      "required_pages_per_second_upper_estimate": 0.005340128242336136
     },
     {
       "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": 9665.553908178752,
-      "required_pages_per_second_upper_estimate": 0.004046765581636881
+      "required_token_positions_per_second": 9648.650510817777,
+      "required_pages_per_second_upper_estimate": 0.004039688482144917
     },
     {
       "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": 19331.107816357504,
-      "required_pages_per_second_upper_estimate": 0.004026159962551882
+      "required_token_positions_per_second": 19297.301021635554,
+      "required_pages_per_second_upper_estimate": 0.00401911889875643
     },
     {
       "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": 28996.661724536258,
-      "required_pages_per_second_upper_estimate": 0.0017383055950471206
+      "required_token_positions_per_second": 28945.95153245333,
+      "required_pages_per_second_upper_estimate": 0.0017352655964617291
     },
     {
       "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": 28996.661724536258,
-      "required_pages_per_second_upper_estimate": 0.00225436290624691
+      "required_token_positions_per_second": 28945.95153245333,
+      "required_pages_per_second_upper_estimate": 0.002250420411863025
     },
     {
       "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": 28996.661724536258,
-      "required_pages_per_second_upper_estimate": 0.37193804241269685
+      "required_token_positions_per_second": 28945.95153245333,
+      "required_pages_per_second_upper_estimate": 0.37128758651701915
     }
   ],
   "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": 0.0921298428594639,
-    "producer_p95_seconds": 0.23165944599895738,
-    "gpu_update_seconds": 0.9534890692815493,
-    "producer_batches_per_second": 10.854246235125066,
-    "gpu_batches_per_second": 1.0487797209395346,
-    "producer_headroom": 10.34940514048216,
+    "producer_batch_seconds": 0.2046149789844094,
+    "producer_p95_seconds": 0.24364660100036417,
+    "gpu_update_seconds": 0.9551594795217526,
+    "producer_batches_per_second": 4.887227733587358,
+    "gpu_batches_per_second": 1.046945584941165,
+    "producer_headroom": 4.668081898317576,
     "sustainable": true,
-    "buffer_seconds": 60.0698113647376,
+    "buffer_seconds": 60.175047209870414,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374123.5418558272,
-    "upload_bytes_per_second": 4755958.498343879,
+    "generation_bytes_per_second": 370787.483183107,
+    "upload_bytes_per_second": 4704717.30724578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.27820293193925,
+    "estimated_full_shards_per_hour": 79.56236240024478,
     "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": 1.2917386739991343,
-      "bytes": 10061132,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 1.2036675580002338,
       "bytes": 326353,
@@ -562,6 +556,12 @@
       "bytes": 6191401,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.1331273229989165,
+      "bytes": 4688984,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 947700, "loss": 0.5153999924659729, "lr": 3e-05, "grad_norm": 0.3330193206786876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138717057800, "tokens_per_second": 153810.06986681506, "data_wait_s": 1.4757999451830983e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 947800, "loss": 0.5304925441741943, "lr": 3e-05, "grad_norm": 0.3326355692209255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138731718809, "tokens_per_second": 152488.53888807967, "data_wait_s": 1.5317000361392274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 947900, "loss": 0.5213189721107483, "lr": 3e-05, "grad_norm": 0.3353032642758652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138746400394, "tokens_per_second": 153313.8390080877, "data_wait_s": 1.2739998055621982e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 948000, "loss": 0.5367710590362549, "lr": 3e-05, "grad_norm": 0.33348746727282125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138760926469, "tokens_per_second": 151720.20473726818, "data_wait_s": 1.1735999578377232e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83d4bd42149ad84456c66fbbc7eaa3dc7165f61c6fe13cd3c6ab792a9f6f66fd
-size 488319
+oid sha256:95ad2a25c5d51e1ae8a4961a9dca7d4da032c0bf7d8f12e38f052e51213620ba
+size 490437