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": 77239.20592701763,
-      "required_pages_per_second_upper_estimate": 0.005343587485680479
+      "required_token_positions_per_second": 77417.88541097565,
+      "required_pages_per_second_upper_estimate": 0.005355948946974218
     },
     {
       "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": 9654.900740877203,
-      "required_pages_per_second_upper_estimate": 0.004042305322951153
+      "required_token_positions_per_second": 9677.235676371956,
+      "required_pages_per_second_upper_estimate": 0.004051656494036316
     },
     {
       "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": 19309.801481754406,
-      "required_pages_per_second_upper_estimate": 0.0040217224149398855
+      "required_token_positions_per_second": 19354.471352743913,
+      "required_pages_per_second_upper_estimate": 0.004031025971043283
     },
     {
       "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": 28964.702222631608,
-      "required_pages_per_second_upper_estimate": 0.0017363896717072706
+      "required_token_positions_per_second": 29031.707029115867,
+      "required_pages_per_second_upper_estimate": 0.0017404065075455864
     },
     {
       "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": 28964.702222631608,
-      "required_pages_per_second_upper_estimate": 0.002251878195548816
+      "required_token_positions_per_second": 29031.707029115867,
+      "required_pages_per_second_upper_estimate": 0.0022570875245299707
     },
     {
       "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": 28964.702222631608,
-      "required_pages_per_second_upper_estimate": 0.37152810023770355
+      "required_token_positions_per_second": 29031.707029115867,
+      "required_pages_per_second_upper_estimate": 0.3723875659511277
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4460233516169865,
-    "producer_p95_seconds": 0.3149598109994258,
-    "gpu_update_seconds": 0.9545411441653696,
-    "producer_batches_per_second": 2.2420350781515355,
-    "gpu_batches_per_second": 1.0476237783069882,
-    "producer_headroom": 2.1401147287576605,
+    "producer_batch_seconds": 0.11158876850777233,
+    "producer_p95_seconds": 0.30262231700180564,
+    "gpu_update_seconds": 0.9523380754797454,
+    "producer_batches_per_second": 8.961475365061927,
+    "gpu_batches_per_second": 1.0500472739118876,
+    "producer_headroom": 8.534354202622225,
     "sustainable": true,
-    "buffer_seconds": 60.13609208241829,
+    "buffer_seconds": 60.9496368307037,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293586.29728157545,
-    "upload_bytes_per_second": 5697703.006294743,
+    "generation_bytes_per_second": 295040.0603261418,
+    "upload_bytes_per_second": 5604018.007299239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.99678505740592,
+    "estimated_full_shards_per_hour": 63.308728764898206,
     "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": 1.3659524169997894,
-      "bytes": 10315345,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.0596161510002275,
       "bytes": 5840926,
@@ -562,6 +556,12 @@
       "bytes": 11892118,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.2482218100012688,
+      "bytes": 2427794,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -312,3 +312,4 @@
 {"step": 676200, "loss": 0.5398682355880737, "lr": 0.0001, "grad_norm": 0.19113540890617703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98978267812, "tokens_per_second": 144062.50475073588, "data_wait_s": 1.7950998881133273e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 676300, "loss": 0.5498566627502441, "lr": 0.0001, "grad_norm": 0.19749079989064325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98992947061, "tokens_per_second": 152402.60035512035, "data_wait_s": 9.996001608669758e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 676400, "loss": 0.5948325395584106, "lr": 0.0001, "grad_norm": 0.20703745292978173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99007715710, "tokens_per_second": 153685.92040189268, "data_wait_s": 1.4428998838411644e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 676500, "loss": 0.5490423440933228, "lr": 0.0001, "grad_norm": 0.20256718652373829, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99022276854, "tokens_per_second": 151011.49398025637, "data_wait_s": 6.705999112455174e-06, "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 sha256:2e472cd3c26a5c4c2e5de799269a87d889e1f3243185b9c771425da39f8a3c82
-size 669915
+oid sha256:2eafd228009cf650bac983b0576cafbe5457da579ee68eaa5aec3afa9e82b88d
+size 672033