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": 76587.28254831888,
-      "required_pages_per_second_upper_estimate": 0.005298485913671456
+      "required_token_positions_per_second": 76286.77510298387,
+      "required_pages_per_second_upper_estimate": 0.005277696111329841
     },
     {
       "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": 9573.41031853986,
-      "required_pages_per_second_upper_estimate": 0.004008186984832034
+      "required_token_positions_per_second": 9535.846887872984,
+      "required_pages_per_second_upper_estimate": 0.003992459960825444
     },
     {
       "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": 19146.82063707972,
-      "required_pages_per_second_upper_estimate": 0.003987777803088057
+      "required_token_positions_per_second": 19071.693775745967,
+      "required_pages_per_second_upper_estimate": 0.003972130859100801
     },
     {
       "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": 28720.230955619576,
-      "required_pages_per_second_upper_estimate": 0.00172173399253591
+      "required_token_positions_per_second": 28607.54066361895,
+      "required_pages_per_second_upper_estimate": 0.0017149783815985854
     },
     {
       "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": 28720.230955619576,
-      "required_pages_per_second_upper_estimate": 0.0022328716298540916
+      "required_token_positions_per_second": 28607.54066361895,
+      "required_pages_per_second_upper_estimate": 0.002224110455323253
     },
     {
       "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": 28720.230955619576,
-      "required_pages_per_second_upper_estimate": 0.3683922853172686
+      "required_token_positions_per_second": 28607.54066361895,
+      "required_pages_per_second_upper_estimate": 0.36694681524889305
     }
   ],
   "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": 0.278069025906305,
-    "producer_p95_seconds": 0.22689579699726892,
-    "gpu_update_seconds": 0.9626663533006938,
-    "producer_batches_per_second": 3.5962293777263374,
-    "gpu_batches_per_second": 1.0387815015776758,
-    "producer_headroom": 3.4619690206886364,
+    "producer_batch_seconds": 0.07897037580457322,
+    "producer_p95_seconds": 0.213879838000139,
+    "gpu_update_seconds": 0.9664584706912878,
+    "producer_batches_per_second": 12.662976335261272,
+    "gpu_batches_per_second": 1.0347056084931623,
+    "producer_headroom": 12.238240743376576,
     "sustainable": true,
-    "buffer_seconds": 60.647980257943715,
+    "buffer_seconds": 60.88688365355113,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418029.72941294033,
-    "upload_bytes_per_second": 1891832.1223800185,
+    "generation_bytes_per_second": 462740.55942897574,
+    "upload_bytes_per_second": 1885423.5114120624,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.6994486979595,
+    "estimated_full_shards_per_hour": 99.29335200454668,
     "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": 1.1467835090006702,
-      "bytes": 6065504,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.5038798329987912,
       "bytes": 7468080,
@@ -562,6 +556,12 @@
       "bytes": 7462921,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.6433775559999049,
+      "bytes": 5520234,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -164,3 +164,4 @@
 {"step": 661400, "loss": 0.5986628532409668, "lr": 0.0001, "grad_norm": 0.20128587924484945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96811615482, "tokens_per_second": 154072.87834906369, "data_wait_s": 8.910001270123757e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661500, "loss": 0.5121641159057617, "lr": 0.0001, "grad_norm": 0.19666728712620124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96826378579, "tokens_per_second": 153557.48604135014, "data_wait_s": 9.351999324280769e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661600, "loss": 0.5818714499473572, "lr": 0.0001, "grad_norm": 0.18780524546508917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96841124320, "tokens_per_second": 153551.00608325418, "data_wait_s": 1.268800042453222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 661700, "loss": 0.5681452751159668, "lr": 0.0001, "grad_norm": 0.17807320642351002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96855865914, "tokens_per_second": 153883.15016461324, "data_wait_s": 1.0070998541777954e-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 sha256:2d2d0afda716ce59deef7923cca9d3ce90ac3a6f3323fad6b30dda3841cc6ca9
-size 354102
+oid sha256:15a8a0a069b3de0d183fb22324600356e7aa2e89dbaef69ff3a5b61f68a6ce87
+size 356220