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": 77207.50810467548,
-      "required_pages_per_second_upper_estimate": 0.0053413945567817684
+      "required_token_positions_per_second": 77131.57731953036,
+      "required_pages_per_second_upper_estimate": 0.0053361414888817345
     },
     {
       "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": 9650.938513084435,
-      "required_pages_per_second_upper_estimate": 0.004040646420915046
+      "required_token_positions_per_second": 9641.447164941295,
+      "required_pages_per_second_upper_estimate": 0.004036672591649401
     },
     {
       "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": 19301.87702616887,
-      "required_pages_per_second_upper_estimate": 0.004020071959823365
+      "required_token_positions_per_second": 19282.89432988259,
+      "required_pages_per_second_upper_estimate": 0.004016118364794299
     },
     {
       "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": 28952.815539253304,
-      "required_pages_per_second_upper_estimate": 0.001735677083188656
+      "required_token_positions_per_second": 28924.341494823886,
+      "required_pages_per_second_upper_estimate": 0.0017339701077024607
     },
     {
       "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": 28952.815539253304,
-      "required_pages_per_second_upper_estimate": 0.0022509540582001477
+      "required_token_positions_per_second": 28924.341494823886,
+      "required_pages_per_second_upper_estimate": 0.0022487403264898455
     },
     {
       "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": 28952.815539253304,
-      "required_pages_per_second_upper_estimate": 0.37137563062625295
+      "required_token_positions_per_second": 28924.341494823886,
+      "required_pages_per_second_upper_estimate": 0.37101039615735926
     }
   ],
   "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": 0.07474459597656846,
-    "producer_p95_seconds": 0.20927418899918848,
-    "gpu_update_seconds": 0.9549330344924736,
-    "producer_batches_per_second": 13.378893643541643,
-    "gpu_batches_per_second": 1.047193849076002,
-    "producer_headroom": 12.775947505179287,
+    "producer_batch_seconds": 0.10796526964075781,
+    "producer_p95_seconds": 0.2406243240002368,
+    "gpu_update_seconds": 0.9558731010331802,
+    "producer_batches_per_second": 9.262237785608155,
+    "gpu_batches_per_second": 1.0461639718903315,
+    "producer_headroom": 8.853523954635964,
     "sustainable": true,
-    "buffer_seconds": 60.16078117302583,
+    "buffer_seconds": 60.22000536509036,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374590.07029887516,
-    "upload_bytes_per_second": 5484937.812296259,
+    "generation_bytes_per_second": 360756.4859132002,
+    "upload_bytes_per_second": 5328368.002377797,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.378309075591,
+    "estimated_full_shards_per_hour": 77.40994389578823,
     "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": 1.0064982160001819,
-      "bytes": 6233391,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.9982179830003588,
       "bytes": 6432855,
@@ -562,6 +556,12 @@
       "bytes": 6319425,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 2.530195315999663,
+      "bytes": 2235467,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -123,3 +123,4 @@
 {"step": 409400, "loss": 0.5289903283119202, "lr": 0.0003, "grad_norm": 0.19719720790661233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59925469566, "tokens_per_second": 151761.8552841852, "data_wait_s": 1.2796999726560898e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 409500, "loss": 0.5733922719955444, "lr": 0.0003, "grad_norm": 0.16575010727183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59940036878, "tokens_per_second": 151550.53328014913, "data_wait_s": 1.3815000784234144e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 409600, "loss": 0.5977052450180054, "lr": 0.0003, "grad_norm": 0.17293401276413004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59954720288, "tokens_per_second": 152388.0654397513, "data_wait_s": 9.227000191458501e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 409700, "loss": 0.6045535206794739, "lr": 0.0003, "grad_norm": 0.1535792509856481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59969279489, "tokens_per_second": 148243.25800603238, "data_wait_s": 1.4062999980524182e-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:c857e78749bf600e29aa669abf2239aea9c7bb9fafee84a4a07716adef3e9f9f
-size 266481
+oid sha256:75c8a46edfe8cf377b74947215cf782ef179af791fc9582203871a1801f7a5d0
+size 268599