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": 77455.83974881544,
-      "required_pages_per_second_upper_estimate": 0.005358574716132171
+      "required_token_positions_per_second": 77107.71522175097,
+      "required_pages_per_second_upper_estimate": 0.0053344906535896615
     },
     {
       "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": 9681.97996860193,
-      "required_pages_per_second_upper_estimate": 0.004053642830120919
+      "required_token_positions_per_second": 9638.464402718871,
+      "required_pages_per_second_upper_estimate": 0.00403542377139403
     },
     {
       "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": 19363.95993720386,
-      "required_pages_per_second_upper_estimate": 0.00403300219295549
+      "required_token_positions_per_second": 19276.928805437743,
+      "required_pages_per_second_upper_estimate": 0.004014875903373896
     },
     {
       "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": 29045.93990580579,
-      "required_pages_per_second_upper_estimate": 0.001741259746770803
+      "required_token_positions_per_second": 28915.393208156613,
+      "required_pages_per_second_upper_estimate": 0.0017334336715800003
     },
     {
       "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": 29045.93990580579,
-      "required_pages_per_second_upper_estimate": 0.002258194068088804
+      "required_token_positions_per_second": 28915.393208156613,
+      "required_pages_per_second_upper_estimate": 0.0022480446365608186
     },
     {
       "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": 29045.93990580579,
-      "required_pages_per_second_upper_estimate": 0.3725701300112337
+      "required_token_positions_per_second": 28915.393208156613,
+      "required_pages_per_second_upper_estimate": 0.37089561714391317
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11651775637497508,
-    "producer_p95_seconds": 0.2340015690024302,
-    "gpu_update_seconds": 0.9518714178181452,
-    "producer_batches_per_second": 8.582382901210527,
-    "gpu_batches_per_second": 1.0505620625653136,
-    "producer_headroom": 8.16932498043347,
+    "producer_batch_seconds": 0.40328230902341033,
+    "producer_p95_seconds": 0.21793054399677203,
+    "gpu_update_seconds": 0.956168909790267,
+    "producer_batches_per_second": 2.479652535271391,
+    "gpu_batches_per_second": 1.0458403214755718,
+    "producer_headroom": 2.3709666613091174,
     "sustainable": true,
-    "buffer_seconds": 60.919770740361294,
+    "buffer_seconds": 60.238641316786826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318946.4076030886,
-    "upload_bytes_per_second": 4464713.332386905,
+    "generation_bytes_per_second": 290697.8459376837,
+    "upload_bytes_per_second": 5108075.817065287,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.43847437924856,
+    "estimated_full_shards_per_hour": 62.376990638712726,
     "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": 14.438800792999245,
-      "bytes": 11778517,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.142588306000107,
       "bytes": 12693549,
@@ -562,6 +556,12 @@
       "bytes": 6205142,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1056236590011395,
+      "bytes": 6648792,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -195,3 +195,4 @@
 {"step": 459700, "loss": 0.577991783618927, "lr": 0.0003, "grad_norm": 0.1984657377628836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67286122725, "tokens_per_second": 154405.87909257287, "data_wait_s": 1.7878002836368978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 459800, "loss": 0.6137932538986206, "lr": 0.0003, "grad_norm": 0.18723392024412436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67300855381, "tokens_per_second": 153362.58875885178, "data_wait_s": 1.6344998584827408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 459900, "loss": 0.6186737418174744, "lr": 0.0003, "grad_norm": 0.1681104369439076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67315439425, "tokens_per_second": 150762.6402912328, "data_wait_s": 1.3208999007474631e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 460000, "loss": 0.599702775478363, "lr": 0.0003, "grad_norm": 0.16255731056963071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67329946830, "tokens_per_second": 150827.70684953185, "data_wait_s": 1.3354001566767693e-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:d7a49ff7fdc7f22b297f6b65419c47842def5b7f90909b9a5312f6c242209914
-size 419760
+oid sha256:0bce7673bd2ae1aad42589fce09f00b633108020b5d66f2bed2d1b60df03db5b
+size 421878