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": 76418.49006065533,
-      "required_pages_per_second_upper_estimate": 0.0052868084577223605
+      "required_token_positions_per_second": 76633.67718640827,
+      "required_pages_per_second_upper_estimate": 0.0053016956023848715
     },
     {
       "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": 9552.311257581916,
-      "required_pages_per_second_upper_estimate": 0.003999353248607438
+      "required_token_positions_per_second": 9579.209648301034,
+      "required_pages_per_second_upper_estimate": 0.0040106150430992615
     },
     {
       "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": 19104.622515163832,
-      "required_pages_per_second_upper_estimate": 0.003978989047132287
+      "required_token_positions_per_second": 19158.419296602067,
+      "required_pages_per_second_upper_estimate": 0.00399019349798934
     },
     {
       "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": 28656.933772745746,
-      "required_pages_per_second_upper_estimate": 0.001717939423071826
+      "required_token_positions_per_second": 28737.6289449031,
+      "required_pages_per_second_upper_estimate": 0.0017227769754282658
     },
     {
       "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": 28656.933772745746,
-      "required_pages_per_second_upper_estimate": 0.002227950552300535
+      "required_token_positions_per_second": 28737.6289449031,
+      "required_pages_per_second_upper_estimate": 0.002234224246995218
     },
     {
       "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": 28656.933772745746,
-      "required_pages_per_second_upper_estimate": 0.367580377018583
+      "required_token_positions_per_second": 28737.6289449031,
+      "required_pages_per_second_upper_estimate": 0.36861544804329216
     }
   ],
   "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": 0.05901574117197583,
-    "producer_p95_seconds": 0.1910415340007603,
-    "gpu_update_seconds": 0.964792682261586,
-    "producer_batches_per_second": 16.94463172267773,
-    "gpu_batches_per_second": 1.0364921069424822,
-    "producer_headroom": 16.348056689657007,
+    "producer_batch_seconds": 0.07765323702348326,
+    "producer_p95_seconds": 0.1996615660009411,
+    "gpu_update_seconds": 0.9620835474286282,
+    "producer_batches_per_second": 12.877763224443408,
+    "gpu_batches_per_second": 1.0394107691298864,
+    "producer_headroom": 12.389484125918445,
     "sustainable": true,
-    "buffer_seconds": 60.781938982479915,
+    "buffer_seconds": 60.61126348800358,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318190.852782727,
-    "upload_bytes_per_second": 4968686.856946572,
+    "generation_bytes_per_second": 336820.76410012186,
+    "upload_bytes_per_second": 5011183.34470542,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.276349903215,
+    "estimated_full_shards_per_hour": 72.27389518978826,
     "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": 0.8067839350001123,
-      "bytes": 1162221,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3463661700002376,
       "bytes": 11675037,
@@ -562,6 +556,12 @@
       "bytes": 7709639,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.0955427600001713,
+      "bytes": 6199869,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -102,3 +102,4 @@
 {"step": 407300, "loss": 0.5221507549285889, "lr": 0.0003, "grad_norm": 0.1611347300725007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59618026986, "tokens_per_second": 151467.20022939064, "data_wait_s": 1.884400080598425e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 407400, "loss": 0.5966902375221252, "lr": 0.0003, "grad_norm": 0.14842046978249865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59632480071, "tokens_per_second": 150283.72701212196, "data_wait_s": 1.2544000128400512e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 407500, "loss": 0.6548508405685425, "lr": 0.0003, "grad_norm": 0.19174734648015654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59647166839, "tokens_per_second": 153215.99214786285, "data_wait_s": 9.211998985847458e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 407600, "loss": 0.6006882190704346, "lr": 0.0003, "grad_norm": 0.16348153630773482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59661852973, "tokens_per_second": 144387.3074612499, "data_wait_s": 1.02319991128752e-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 sha256:9ba1fd01032498aa7336f1d7b09993c6eecc8ea30d1a730131f3ec40f773e647
-size 222003
+oid sha256:cadcf369aad5fb8374ce94aa6f0d4aae47ddbe194b02a49ddfdeac1dc61615e1
+size 224121