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": 77159.85173997945,
-      "required_pages_per_second_upper_estimate": 0.005338097578894097
+      "required_token_positions_per_second": 76360.01345607374,
+      "required_pages_per_second_upper_estimate": 0.00528276291053299
     },
     {
       "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": 9644.981467497431,
-      "required_pages_per_second_upper_estimate": 0.004038152330325026
+      "required_token_positions_per_second": 9545.001682009217,
+      "required_pages_per_second_upper_estimate": 0.003996292881956445
     },
     {
       "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": 19289.962934994863,
-      "required_pages_per_second_upper_estimate": 0.0040175905688275325
+      "required_token_positions_per_second": 19090.003364018434,
+      "required_pages_per_second_upper_estimate": 0.003975944263481642
     },
     {
       "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": 28934.944402492296,
-      "required_pages_per_second_upper_estimate": 0.0017346057358274794
+      "required_token_positions_per_second": 28635.00504602765,
+      "required_pages_per_second_upper_estimate": 0.0017166248293883032
     },
     {
       "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": 28934.944402492296,
-      "required_pages_per_second_upper_estimate": 0.0022495646559238026
+      "required_token_positions_per_second": 28635.00504602765,
+      "required_pages_per_second_upper_estimate": 0.002226245690252478
     },
     {
       "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": 28934.944402492296,
-      "required_pages_per_second_upper_estimate": 0.37114639887241435
+      "required_token_positions_per_second": 28635.00504602765,
+      "required_pages_per_second_upper_estimate": 0.3672990988574755
     }
   ],
   "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": 0.10032429012466082,
-    "producer_p95_seconds": 0.29078306700102985,
-    "gpu_update_seconds": 0.955522831335337,
-    "producer_batches_per_second": 9.967675811684503,
-    "gpu_batches_per_second": 1.0465474682614402,
-    "producer_headroom": 9.524341813413528,
+    "producer_batch_seconds": 0.4100210520155372,
+    "producer_p95_seconds": 0.2947956939970027,
+    "gpu_update_seconds": 0.9655315218404485,
+    "producer_batches_per_second": 2.4388991616023326,
+    "gpu_batches_per_second": 1.0356989672319028,
+    "producer_headroom": 2.3548340191172943,
     "sustainable": true,
-    "buffer_seconds": 60.19793837412623,
+    "buffer_seconds": 60.82848587594826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402124.67177934566,
+    "generation_bytes_per_second": 343453.55065844127,
     "upload_bytes_per_second": 4719464.787385518,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.2865935805824,
+    "estimated_full_shards_per_hour": 73.69713678183487,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 870400, "loss": 0.5406244993209839, "lr": 3e-05, "grad_norm": 0.3112898802282261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127402314772, "tokens_per_second": 153446.15395961766, "data_wait_s": 9.04400076251477e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 870500, "loss": 0.4992218315601349, "lr": 3e-05, "grad_norm": 0.3150675087006588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127417027799, "tokens_per_second": 152484.79115875156, "data_wait_s": 1.0513002052903175e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870600, "loss": 0.5031870603561401, "lr": 3e-05, "grad_norm": 0.30149928247060004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127431503243, "tokens_per_second": 149693.02054720913, "data_wait_s": 1.001199780148454e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 870700, "loss": 0.520969808101654, "lr": 3e-05, "grad_norm": 0.30597625598022377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127446027794, "tokens_per_second": 150416.08419461123, "data_wait_s": 1.535900082672015e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af8f82ec051e5e5746800d0122cb483b7ee2702ec4a147e8bc40186e3a1d7652
-size 600021
+oid sha256:e26060b3f7b3bb3936e124d2524bbd1fd4e15e7dbf2ef0e75c5e0831529f42ff
+size 602139