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": 77036.10101510357,
-      "required_pages_per_second_upper_estimate": 0.005329536216605941
+      "required_token_positions_per_second": 76550.91007371704,
+      "required_pages_per_second_upper_estimate": 0.005295969581482732
     },
     {
       "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": 9629.512626887947,
-      "required_pages_per_second_upper_estimate": 0.004031675849787959
+      "required_token_positions_per_second": 9568.86375921463,
+      "required_pages_per_second_upper_estimate": 0.004006283435385515
     },
     {
       "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": 19259.025253775893,
-      "required_pages_per_second_upper_estimate": 0.004011147065711141
+      "required_token_positions_per_second": 19137.72751842926,
+      "required_pages_per_second_upper_estimate": 0.00398588394627483
     },
     {
       "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": 28888.53788066384,
-      "required_pages_per_second_upper_estimate": 0.0017318237357025195
+      "required_token_positions_per_second": 28706.591277643893,
+      "required_pages_per_second_upper_estimate": 0.0017209163146678502
     },
     {
       "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": 28888.53788066384,
-      "required_pages_per_second_upper_estimate": 0.0022459567529723575
+      "required_token_positions_per_second": 28706.591277643893,
+      "required_pages_per_second_upper_estimate": 0.002231811204886085
     },
     {
       "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": 28888.53788066384,
-      "required_pages_per_second_upper_estimate": 0.37055114583783993
+      "required_token_positions_per_second": 28706.591277643893,
+      "required_pages_per_second_upper_estimate": 0.3682173301733417
     }
   ],
   "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": 0.10787594935129619,
-    "producer_p95_seconds": 0.3014711770010763,
-    "gpu_update_seconds": 0.9570577823706967,
-    "producer_batches_per_second": 9.269906832926374,
-    "gpu_batches_per_second": 1.0448689916328067,
-    "producer_headroom": 8.871836476303484,
+    "producer_batch_seconds": 0.28332189247612405,
+    "producer_p95_seconds": 0.30531923600210575,
+    "gpu_update_seconds": 0.9631237555373459,
+    "producer_batches_per_second": 3.5295542863291844,
+    "gpu_batches_per_second": 1.0382881683175598,
+    "producer_headroom": 3.399397579622301,
     "sustainable": true,
-    "buffer_seconds": 60.29464028935389,
+    "buffer_seconds": 60.67679659885279,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516398.01263888687,
-    "upload_bytes_per_second": 5338976.7067648005,
+    "generation_bytes_per_second": 486837.46028524067,
+    "upload_bytes_per_second": 5407975.678774594,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.8069923818107,
+    "estimated_full_shards_per_hour": 104.4639859811584,
     "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": 1.0325014549998741,
-      "bytes": 941600,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.327213976001076,
       "bytes": 13471794,
@@ -562,6 +556,12 @@
       "bytes": 5972281,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2648427659987647,
+      "bytes": 7742666,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"step": 672000, "loss": 0.5388064980506897, "lr": 0.0001, "grad_norm": 0.18466606216347412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98363623643, "tokens_per_second": 150813.16569094366, "data_wait_s": 1.1814998288173229e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 672100, "loss": 0.5250399112701416, "lr": 0.0001, "grad_norm": 0.21554353769799287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98378141448, "tokens_per_second": 151300.95974596552, "data_wait_s": 8.882001566234976e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 672200, "loss": 0.5884952545166016, "lr": 0.0001, "grad_norm": 0.19908141212471916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98392759379, "tokens_per_second": 151470.59718797472, "data_wait_s": 1.1619002179941162e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 672300, "loss": 0.5632578730583191, "lr": 0.0001, "grad_norm": 0.21088603035231673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98407503724, "tokens_per_second": 153502.58779994672, "data_wait_s": 1.6398000298067927e-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 sha256:72c79c430af409ad5b707bb8272a01be5d6b098fb5cbccef6e2b75a3ddef8b26
-size 580176
+oid sha256:e668c2928119f2566eb5b4960c9c0ab44c0683cd7d19a6721ea14aa8b029a16e
+size 582294