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": 76334.93662462175,
-      "required_pages_per_second_upper_estimate": 0.005281028037146877
+      "required_token_positions_per_second": 76867.87701115926,
+      "required_pages_per_second_upper_estimate": 0.005317898089679607
     },
     {
       "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": 9541.86707807772,
-      "required_pages_per_second_upper_estimate": 0.0039949804887483
+      "required_token_positions_per_second": 9608.484626394908,
+      "required_pages_per_second_upper_estimate": 0.004022871865096107
     },
     {
       "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": 19083.73415615544,
-      "required_pages_per_second_upper_estimate": 0.00397463855281391
+      "required_token_positions_per_second": 19216.969252789815,
+      "required_pages_per_second_upper_estimate": 0.00400238790979707
     },
     {
       "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": 28625.60123423316,
-      "required_pages_per_second_upper_estimate": 0.0017160610852300128
+      "required_token_positions_per_second": 28825.453879184723,
+      "required_pages_per_second_upper_estimate": 0.0017280419461375392
     },
     {
       "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": 28625.60123423316,
-      "required_pages_per_second_upper_estimate": 0.002225514585248453
+      "required_token_positions_per_second": 28825.453879184723,
+      "required_pages_per_second_upper_estimate": 0.0022410522493345538
     },
     {
       "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": 28625.60123423316,
-      "required_pages_per_second_upper_estimate": 0.3671784768568022
+      "required_token_positions_per_second": 28825.453879184723,
+      "required_pages_per_second_upper_estimate": 0.36974197200118936
     }
   ],
   "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": 0.3901215416877335,
-    "producer_p95_seconds": 0.3408100980013842,
-    "gpu_update_seconds": 0.9658487091246121,
-    "producer_batches_per_second": 2.5633037224087305,
-    "gpu_batches_per_second": 1.0353588409372527,
-    "producer_headroom": 2.475763591382785,
+    "producer_batch_seconds": 0.13822678385162135,
+    "producer_p95_seconds": 0.33734097300475696,
+    "gpu_update_seconds": 0.9591522865825548,
+    "producer_batches_per_second": 7.234487934505107,
+    "gpu_batches_per_second": 1.0425873075515306,
+    "producer_headroom": 6.9389756446344775,
     "sustainable": true,
-    "buffer_seconds": 60.84846867485056,
+    "buffer_seconds": 60.42659405470095,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404012.84595853544,
-    "upload_bytes_per_second": 5177145.724026455,
+    "generation_bytes_per_second": 410057.85212746606,
+    "upload_bytes_per_second": 5223682.262969395,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.69175180499123,
+    "estimated_full_shards_per_hour": 87.98886940830218,
     "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": 0.8352084129983268,
-      "bytes": 2708130,
-      "pages": 39,
-      "shards": 2
-    },
     {
       "seconds": 0.9732167729998764,
       "bytes": 6207933,
@@ -562,6 +556,12 @@
       "bytes": 6981210,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0006479639996542,
+      "bytes": 7455931,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 435300, "loss": 0.5720965266227722, "lr": 0.0003, "grad_norm": 0.19244354152391166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63715228629, "tokens_per_second": 151695.70800136589, "data_wait_s": 1.065700053004548e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435400, "loss": 0.5408660173416138, "lr": 0.0003, "grad_norm": 0.1576852767167392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63729851280, "tokens_per_second": 151528.96290027507, "data_wait_s": 1.2777003576047719e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435500, "loss": 0.5721021890640259, "lr": 0.0003, "grad_norm": 0.15697470459950236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63744483124, "tokens_per_second": 152018.54505581406, "data_wait_s": 1.4756995369680226e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 435600, "loss": 0.5836086273193359, "lr": 0.0003, "grad_norm": 0.15119505337384306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63759238878, "tokens_per_second": 152342.34353668164, "data_wait_s": 1.897000038297847e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:320e42ea5b2fbb6f5c595330c32ab5301fbb4ae1b2091284ff73e71dcc767317
-size 819741
+oid sha256:d45bcb492aed956f8b4d19fd5581c9a0ab424c090c47b79c009378d99315632d
+size 821859