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": 75102.05142626633,
-      "required_pages_per_second_upper_estimate": 0.005195734178436862
+      "required_token_positions_per_second": 76093.02896483314,
+      "required_pages_per_second_upper_estimate": 0.005264292303939597
     },
     {
       "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": 9387.756428283292,
-      "required_pages_per_second_upper_estimate": 0.003930457577875694
+      "required_token_positions_per_second": 9511.628620604142,
+      "required_pages_per_second_upper_estimate": 0.00398232027805491
     },
     {
       "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": 18775.512856566584,
-      "required_pages_per_second_upper_estimate": 0.003910444184451829
+      "required_token_positions_per_second": 19023.257241208285,
+      "required_pages_per_second_upper_estimate": 0.003962042806313916
     },
     {
       "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": 28163.269284849877,
-      "required_pages_per_second_upper_estimate": 0.0016883449908044964
+      "required_token_positions_per_second": 28534.88586181243,
+      "required_pages_per_second_upper_estimate": 0.0017106228371677433
     },
     {
       "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": 28163.269284849877,
-      "required_pages_per_second_upper_estimate": 0.0021895703097672328
+      "required_token_positions_per_second": 28534.88586181243,
+      "required_pages_per_second_upper_estimate": 0.0022184618640574954
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28163.269284849877,
-      "required_pages_per_second_upper_estimate": 0.36124817902348455
+      "required_token_positions_per_second": 28534.88586181243,
+      "required_pages_per_second_upper_estimate": 0.36601487746196726
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10015173966397128,
-    "producer_p95_seconds": 0.2521660480015271,
-    "gpu_update_seconds": 0.9817042091371452,
-    "producer_batches_per_second": 9.984849023643484,
-    "gpu_batches_per_second": 1.0186367652217114,
-    "producer_headroom": 9.802168314109721,
+    "producer_batch_seconds": 0.5729266434686906,
+    "producer_p95_seconds": 0.2820500230009202,
+    "gpu_update_seconds": 0.9689192427084726,
+    "producer_batches_per_second": 1.7454241505433639,
+    "gpu_batches_per_second": 1.0320777583120815,
+    "producer_headroom": 1.6911750461495554,
     "sustainable": true,
-    "buffer_seconds": 60.865660966503,
+    "buffer_seconds": 60.0729930479253,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293232.6851324216,
-    "upload_bytes_per_second": 4969374.818953827,
+    "generation_bytes_per_second": 274704.74983620754,
+    "upload_bytes_per_second": 5046865.861083235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.92090812186705,
+    "estimated_full_shards_per_hour": 58.945244515558905,
     "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": 1.210031186001288,
-      "bytes": 5583238,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.2692713039996306,
       "bytes": 11755459,
@@ -562,6 +556,12 @@
       "bytes": 6189631,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3973069029998442,
+      "bytes": 13145960,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 736900, "loss": 0.49935975670814514, "lr": 5e-05, "grad_norm": 0.2256634917099695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107861781751, "tokens_per_second": 148566.06792430306, "data_wait_s": 1.6441998013760895e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737000, "loss": 0.5023998022079468, "lr": 5e-05, "grad_norm": 0.22182048485091402, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107876384978, "tokens_per_second": 149998.93509704448, "data_wait_s": 1.329600127064623e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737100, "loss": 0.5295858979225159, "lr": 5e-05, "grad_norm": 0.22646088657728392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107890989482, "tokens_per_second": 150112.78568658954, "data_wait_s": 1.0529998689889908e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 737200, "loss": 0.5161051154136658, "lr": 5e-05, "grad_norm": 0.2209857690773546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107905571719, "tokens_per_second": 148907.93041877376, "data_wait_s": 1.9220999092794955e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8aa9b99fab756b0c76ca011a2c10b9f967430b7c65d9003d3e15a84a0cf4b3c1
-size 364692
+oid sha256:2456f26e902f7a8939c8761d42bb0d51622dea62277e4d6119d58b3c5ac79e7d
+size 366810