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": 77248.17804031298,
-      "required_pages_per_second_upper_estimate": 0.005344208197296423
+      "required_token_positions_per_second": 76919.94777012474,
+      "required_pages_per_second_upper_estimate": 0.005321500465605645
     },
     {
       "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": 9656.022255039123,
-      "required_pages_per_second_upper_estimate": 0.00404277487751087
+      "required_token_positions_per_second": 9614.993471265592,
+      "required_pages_per_second_upper_estimate": 0.004025596982523326
     },
     {
       "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": 19312.044510078245,
-      "required_pages_per_second_upper_estimate": 0.004022189578587106
+      "required_token_positions_per_second": 19229.986942531184,
+      "required_pages_per_second_upper_estimate": 0.004005099151270632
     },
     {
       "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": 28968.06676511737,
-      "required_pages_per_second_upper_estimate": 0.0017365913708919333
+      "required_token_positions_per_second": 28844.980413796777,
+      "required_pages_per_second_upper_estimate": 0.00172921253207224
     },
     {
       "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": 28968.06676511737,
-      "required_pages_per_second_upper_estimate": 0.002252139773928026
+      "required_token_positions_per_second": 28844.980413796777,
+      "required_pages_per_second_upper_estimate": 0.002242570351512492
     },
     {
       "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": 28968.06676511737,
-      "required_pages_per_second_upper_estimate": 0.3715712569761467
+      "required_token_positions_per_second": 28844.980413796777,
+      "required_pages_per_second_upper_estimate": 0.3699924374212334
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10852802986718757,
-    "producer_p95_seconds": 0.2601833090011496,
-    "gpu_update_seconds": 0.9544302774561759,
-    "producer_batches_per_second": 9.214209464815324,
-    "gpu_batches_per_second": 1.0477454703818494,
-    "producer_headroom": 8.794320496043014,
+    "producer_batch_seconds": 0.08993021281250435,
+    "producer_p95_seconds": 0.26975980500355945,
+    "gpu_update_seconds": 0.9585029909320288,
+    "producer_batches_per_second": 11.119733499184548,
+    "gpu_batches_per_second": 1.0432935624203117,
+    "producer_headroom": 10.658297817335463,
     "sustainable": true,
-    "buffer_seconds": 60.12910747973908,
+    "buffer_seconds": 60.38568842871782,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 507057.5429060559,
+    "generation_bytes_per_second": 505934.9399003648,
     "upload_bytes_per_second": 4787220.251045446,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.80274501215227,
+    "estimated_full_shards_per_hour": 108.56186053999146,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -244,3 +244,4 @@
 {"step": 464600, "loss": 0.6125659942626953, "lr": 0.0003, "grad_norm": 0.1611607892049086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68003346812, "tokens_per_second": 152839.153010274, "data_wait_s": 1.1713000276358798e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464700, "loss": 0.5485379695892334, "lr": 0.0003, "grad_norm": 0.1977374657831521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68017987997, "tokens_per_second": 152362.4185223103, "data_wait_s": 1.6730002244003117e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464800, "loss": 0.5927941799163818, "lr": 0.0003, "grad_norm": 0.1538716172456637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68032688293, "tokens_per_second": 152428.54795846582, "data_wait_s": 8.838000212563202e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 464900, "loss": 0.5823112726211548, "lr": 0.0003, "grad_norm": 0.1557854975686474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68047434277, "tokens_per_second": 153518.83346817302, "data_wait_s": 1.333300315309316e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6996d7f7263a8e58e32e6165e10dbabd80b338430523f07eb43c0b53fcb5e97c
-size 524325
+oid sha256:ef2dced8eeef27494145de6aea9815fcbf523045b45ea8cbe9e35f5bc90bfdb2
+size 526443