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": 76576.99136215974,
-      "required_pages_per_second_upper_estimate": 0.0052977739455341245
+      "required_token_positions_per_second": 77327.49932541247,
+      "required_pages_per_second_upper_estimate": 0.0053496958278503445
     },
     {
       "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": 9572.123920269967,
-      "required_pages_per_second_upper_estimate": 0.004007648396739481
+      "required_token_positions_per_second": 9665.937415676559,
+      "required_pages_per_second_upper_estimate": 0.004046926148217666
     },
     {
       "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": 19144.247840539934,
-      "required_pages_per_second_upper_estimate": 0.003987241957418027
+      "required_token_positions_per_second": 19331.874831353118,
+      "required_pages_per_second_upper_estimate": 0.004026319711547923
     },
     {
       "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": 28716.3717608099,
-      "required_pages_per_second_upper_estimate": 0.0017215026396997192
+      "required_token_positions_per_second": 28997.812247029673,
+      "required_pages_per_second_upper_estimate": 0.0017383745671138548
     },
     {
       "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": 28716.3717608099,
-      "required_pages_per_second_upper_estimate": 0.002232571594432443
+      "required_token_positions_per_second": 28997.812247029673,
+      "required_pages_per_second_upper_estimate": 0.0022544523543101594
     },
     {
       "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": 28716.3717608099,
-      "required_pages_per_second_upper_estimate": 0.3683427837099306
+      "required_token_positions_per_second": 28997.812247029673,
+      "required_pages_per_second_upper_estimate": 0.37195280007990755
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06125748567963285,
-    "producer_p95_seconds": 0.2088843330002419,
-    "gpu_update_seconds": 0.9627957260858442,
-    "producer_batches_per_second": 16.324535506237474,
-    "gpu_batches_per_second": 1.038641918432071,
-    "producer_headroom": 15.717193015742051,
+    "producer_batch_seconds": 0.19436496217963395,
+    "producer_p95_seconds": 0.2066818390012486,
+    "gpu_update_seconds": 0.9534512384751391,
+    "producer_batches_per_second": 5.144960227326314,
+    "gpu_batches_per_second": 1.048821334166293,
+    "producer_headroom": 4.905468700649608,
     "sustainable": true,
-    "buffer_seconds": 60.656130743408184,
+    "buffer_seconds": 60.067428023933765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342310.9154179084,
+    "generation_bytes_per_second": 351507.3999616873,
     "upload_bytes_per_second": 3539218.4148892364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.45195385840358,
+    "estimated_full_shards_per_hour": 75.42530535829509,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 940300, "loss": 0.5543428063392639, "lr": 3e-05, "grad_norm": 0.34451358979361424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137633239476, "tokens_per_second": 153214.84122930956, "data_wait_s": 8.251001418102533e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 940400, "loss": 0.5169722437858582, "lr": 3e-05, "grad_norm": 0.33658028024018627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137647828212, "tokens_per_second": 151341.28777122946, "data_wait_s": 1.2767999578500167e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940500, "loss": 0.5382846593856812, "lr": 3e-05, "grad_norm": 0.3356429533463776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137662492808, "tokens_per_second": 153278.8204821817, "data_wait_s": 1.438699837308377e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 940600, "loss": 0.5275529623031616, "lr": 3e-05, "grad_norm": 0.3333261844303751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137677145445, "tokens_per_second": 152856.9717985757, "data_wait_s": 1.476199940952938e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62ba45edda30e5125470ecca7bbaabc84f15e2443753ffdd0878a92c52664031
-size 330804
+oid sha256:25523570a5baf6139fa7c100051253220eb611a756f44d8b4ceb3a98e5725ab9
+size 332922