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": 77333.84104640863,
-      "required_pages_per_second_upper_estimate": 0.005350134562823679
+      "required_token_positions_per_second": 76512.91576159814,
+      "required_pages_per_second_upper_estimate": 0.005293341046811397
     },
     {
       "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": 9666.73013080108,
-      "required_pages_per_second_upper_estimate": 0.004047258041486528
+      "required_token_positions_per_second": 9564.114470199767,
+      "required_pages_per_second_upper_estimate": 0.004004295007251425
     },
     {
       "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": 19333.46026160216,
-      "required_pages_per_second_upper_estimate": 0.00402664991485819
+      "required_token_positions_per_second": 19128.228940399535,
+      "required_pages_per_second_upper_estimate": 0.003983905642965586
     },
     {
       "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": 29000.190392403238,
-      "required_pages_per_second_upper_estimate": 0.001738517133297782
+      "required_token_positions_per_second": 28692.3434105993,
+      "required_pages_per_second_upper_estimate": 0.0017200621767937603
     },
     {
       "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": 29000.190392403238,
-      "required_pages_per_second_upper_estimate": 0.0022546372446525984
+      "required_token_positions_per_second": 28692.3434105993,
+      "required_pages_per_second_upper_estimate": 0.002230703495893112
     },
     {
       "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": 29000.190392403238,
-      "required_pages_per_second_upper_estimate": 0.3719833043753061
+      "required_token_positions_per_second": 28692.3434105993,
+      "required_pages_per_second_upper_estimate": 0.36803457383306143
     }
   ],
   "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": 0.0780217244452075,
-    "producer_p95_seconds": 0.20309704200008127,
-    "gpu_update_seconds": 0.9533730512073655,
-    "producer_batches_per_second": 12.816943064393206,
-    "gpu_batches_per_second": 1.0489073492622698,
-    "producer_headroom": 12.219328116451631,
+    "producer_batch_seconds": 0.4920465686406459,
+    "producer_p95_seconds": 0.22106794200044533,
+    "gpu_update_seconds": 0.9636020175956242,
+    "producer_batches_per_second": 2.032327961889163,
+    "gpu_batches_per_second": 1.0377728374782733,
+    "producer_headroom": 1.9583553244924004,
     "sustainable": true,
-    "buffer_seconds": 60.062502226064026,
+    "buffer_seconds": 60.706927108524326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331352.40544262616,
-    "upload_bytes_per_second": 5532668.14553594,
+    "generation_bytes_per_second": 304849.90521359997,
+    "upload_bytes_per_second": 5498385.0564517835,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.1005127187642,
+    "estimated_full_shards_per_hour": 65.41369311624526,
     "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": 1.497052147999966,
-      "bytes": 11817101,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.0985135669998272,
       "bytes": 6249223,
@@ -562,6 +556,12 @@
       "bytes": 7321171,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0708932610014017,
+      "bytes": 6812845,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -114,3 +114,4 @@
 {"step": 771400, "loss": 0.5066722631454468, "lr": 3e-05, "grad_norm": 0.24753382743820754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112910503453, "tokens_per_second": 152183.82168502882, "data_wait_s": 1.6093999875010923e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 771500, "loss": 0.5179535150527954, "lr": 3e-05, "grad_norm": 0.24759907311140444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112925242300, "tokens_per_second": 153496.88577933644, "data_wait_s": 1.0863999705179594e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 771600, "loss": 0.547562301158905, "lr": 3e-05, "grad_norm": 0.25522751868892374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112939946018, "tokens_per_second": 137361.2615154882, "data_wait_s": 2.446300095471088e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 771700, "loss": 0.5201507806777954, "lr": 3e-05, "grad_norm": 0.2466334159395461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112954677935, "tokens_per_second": 153370.7300342805, "data_wait_s": 1.5510999219259247e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cff39271e6f20c20814ad107432eb55b64fc020a2ec20901fa57579d1877654a
-size 247419
+oid sha256:3c61a1dbb35af58893a58fa7a2b20eba4d0820e7a1862bd5eff3345c905a82df
+size 249537