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": 76150.50956321313,
-      "required_pages_per_second_upper_estimate": 0.005268268945108888
+      "required_token_positions_per_second": 76183.11933318166,
+      "required_pages_per_second_upper_estimate": 0.005270524964660402
     },
     {
       "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": 9518.81369540164,
-      "required_pages_per_second_upper_estimate": 0.003985328518831199
+      "required_token_positions_per_second": 9522.889916647708,
+      "required_pages_per_second_upper_estimate": 0.003987035147545748
     },
     {
       "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": 19037.62739080328,
-      "required_pages_per_second_upper_estimate": 0.003965035729508225
+      "required_token_positions_per_second": 19045.779833295415,
+      "required_pages_per_second_upper_estimate": 0.00396673366828497
     },
     {
       "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": 28556.441086204923,
-      "required_pages_per_second_upper_estimate": 0.0017119150399571514
+      "required_token_positions_per_second": 28568.669749943125,
+      "required_pages_per_second_upper_estimate": 0.0017126481296761682
     },
     {
       "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": 28556.441086204923,
-      "required_pages_per_second_upper_estimate": 0.0022201376879425997
+      "required_token_positions_per_second": 28568.669749943125,
+      "required_pages_per_second_upper_estimate": 0.0022210884127600375
     },
     {
       "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": 28556.441086204923,
-      "required_pages_per_second_upper_estimate": 0.3662913647362774
+      "required_token_positions_per_second": 28568.669749943125,
+      "required_pages_per_second_upper_estimate": 0.36644822090459495
     }
   ],
   "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": 0.10760444438295735,
-    "producer_p95_seconds": 0.3218163010024,
-    "gpu_update_seconds": 0.9681878745512243,
-    "producer_batches_per_second": 9.293296440814878,
-    "gpu_batches_per_second": 1.0328573888239627,
-    "producer_headroom": 8.997656928607014,
+    "producer_batch_seconds": 0.08709490440617174,
+    "producer_p95_seconds": 0.32369011800255976,
+    "gpu_update_seconds": 0.9677734469962517,
+    "producer_batches_per_second": 11.481727970403947,
+    "gpu_batches_per_second": 1.033299687136253,
+    "producer_headroom": 11.111711455391104,
     "sustainable": true,
-    "buffer_seconds": 60.02764822217591,
+    "buffer_seconds": 60.001953713767605,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323056.80236905173,
-    "upload_bytes_per_second": 5254465.453152432,
+    "generation_bytes_per_second": 363341.0464285099,
+    "upload_bytes_per_second": 5278377.225503922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.32046941093124,
+    "estimated_full_shards_per_hour": 77.96453041688416,
     "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": 0.9163747099992179,
-      "bytes": 2165474,
-      "pages": 22,
-      "shards": 2
-    },
     {
       "seconds": 1.0913281839966658,
       "bytes": 12193503,
@@ -562,6 +556,12 @@
       "bytes": 2428421,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.3105249759973958,
+      "bytes": 6190634,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 637400, "loss": 0.5920100212097168, "lr": 0.0001, "grad_norm": 0.1860952955176649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93298632099, "tokens_per_second": 150522.39252012363, "data_wait_s": 9.453000529902056e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 637500, "loss": 0.569724440574646, "lr": 0.0001, "grad_norm": 0.17819412745127783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93313262112, "tokens_per_second": 152721.54016342905, "data_wait_s": 1.1635002010734752e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637600, "loss": 0.5447022914886475, "lr": 0.0001, "grad_norm": 0.21281934312340497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93328093606, "tokens_per_second": 152960.06271889122, "data_wait_s": 1.8478000129107386e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 637700, "loss": 0.5084192156791687, "lr": 0.0001, "grad_norm": 0.171511491103215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93342791224, "tokens_per_second": 152764.1056242407, "data_wait_s": 1.5092999092303216e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0c46f68f04f2b12ad7a41c31bb84ccfda73609c46b8903b3c36edc76061490b
-size 708822
+oid sha256:913999db802f654b0983e408805baeb828fefa9cab41714115b7d9b713ae6d8c
+size 710940