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": 76394.06528418539,
-      "required_pages_per_second_upper_estimate": 0.005285118695012881
+      "required_token_positions_per_second": 76443.91829991365,
+      "required_pages_per_second_upper_estimate": 0.005288567642315896
     },
     {
       "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": 9549.258160523174,
-      "required_pages_per_second_upper_estimate": 0.003998074980624859
+      "required_token_positions_per_second": 9555.489787489207,
+      "required_pages_per_second_upper_estimate": 0.004000684032704369
     },
     {
       "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": 19098.516321046347,
-      "required_pages_per_second_upper_estimate": 0.003977717287928783
+      "required_token_positions_per_second": 19110.979574978413,
+      "required_pages_per_second_upper_estimate": 0.003980313055044723
     },
     {
       "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": 28647.774481569522,
-      "required_pages_per_second_upper_estimate": 0.0017173903375512392
+      "required_token_positions_per_second": 28666.46936246762,
+      "required_pages_per_second_upper_estimate": 0.0017185110671156499
     },
     {
       "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": 28647.774481569522,
-      "required_pages_per_second_upper_estimate": 0.0022272384576990474
+      "required_token_positions_per_second": 28666.46936246762,
+      "required_pages_per_second_upper_estimate": 0.002228691902458784
     },
     {
       "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": 28647.774481569522,
-      "required_pages_per_second_upper_estimate": 0.36746289146585503
+      "required_token_positions_per_second": 28666.46936246762,
+      "required_pages_per_second_upper_estimate": 0.3677026893250166
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.13499754862499458,
-    "producer_p95_seconds": 0.3564191400000709,
-    "gpu_update_seconds": 0.9651011466104383,
-    "producer_batches_per_second": 7.407541916022996,
-    "gpu_batches_per_second": 1.0361608247095457,
-    "producer_headroom": 7.149027196718677,
+    "producer_batch_seconds": 0.4768454809221794,
+    "producer_p95_seconds": 0.34351110600255197,
+    "gpu_update_seconds": 0.964471754453268,
+    "producer_batches_per_second": 2.0971153969333702,
+    "gpu_batches_per_second": 1.0368369995105475,
+    "producer_headroom": 2.0226085661712894,
     "sustainable": true,
-    "buffer_seconds": 60.80137223645762,
+    "buffer_seconds": 60.761720530555884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363408.05796621257,
-    "upload_bytes_per_second": 5138219.861428762,
+    "generation_bytes_per_second": 330206.4953332178,
+    "upload_bytes_per_second": 5179234.674758404,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.9789095329264,
+    "estimated_full_shards_per_hour": 70.85462708470726,
     "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": 1.2169118900019384,
-      "bytes": 10000231,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.287032372001704,
       "bytes": 6056411,
@@ -562,6 +556,12 @@
       "bytes": 10179020,
       "pages": 46,
       "shards": 3
+    },
+    {
+      "seconds": 1.3179188790018088,
+      "bytes": 13975639,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 436700, "loss": 0.5688666701316833, "lr": 0.0003, "grad_norm": 0.14958992454250414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63920360215, "tokens_per_second": 151443.07067067074, "data_wait_s": 9.973002306651324e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 436800, "loss": 0.6163980960845947, "lr": 0.0003, "grad_norm": 0.1846098944574017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63934942003, "tokens_per_second": 150709.02231486028, "data_wait_s": 1.6042999050114304e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436900, "loss": 0.6994632482528687, "lr": 0.0003, "grad_norm": 0.16904049871297688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63949527615, "tokens_per_second": 151376.60636436543, "data_wait_s": 1.4575998648069799e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 437000, "loss": 0.5981572866439819, "lr": 0.0003, "grad_norm": 0.14569365186351016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63964172030, "tokens_per_second": 152071.62493948467, "data_wait_s": 1.4822995581198484e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c12d5c3c4bf442a209db8348f9bc37adbf6e88d70d35b13c822fda0601783640
-size 849393
+oid sha256:bec1a1b031e60d7351fa956e0447625cfaf70c792e216832707e542af88b02e6
+size 851511