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": 77292.46244246345,
-      "required_pages_per_second_upper_estimate": 0.005347271894991163
+      "required_token_positions_per_second": 76835.46975774648,
+      "required_pages_per_second_upper_estimate": 0.005315656080693323
     },
     {
       "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": 9661.557805307932,
-      "required_pages_per_second_upper_estimate": 0.004045092496812986
+      "required_token_positions_per_second": 9604.43371971831,
+      "required_pages_per_second_upper_estimate": 0.004021175835063166
     },
     {
       "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": 19323.115610615863,
-      "required_pages_per_second_upper_estimate": 0.004024495396864537
+      "required_token_positions_per_second": 19208.86743943662,
+      "required_pages_per_second_upper_estimate": 0.004000700515734785
     },
     {
       "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": 28984.673415923793,
-      "required_pages_per_second_upper_estimate": 0.0017375869142508907
+      "required_token_positions_per_second": 28813.30115915493,
+      "required_pages_per_second_upper_estimate": 0.0017273134091278822
     },
     {
       "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": 28984.673415923793,
-      "required_pages_per_second_upper_estimate": 0.002253430867983288
+      "required_token_positions_per_second": 28813.30115915493,
+      "required_pages_per_second_upper_estimate": 0.0022401074288064034
     },
     {
       "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": 28984.673415923793,
-      "required_pages_per_second_upper_estimate": 0.37178426926185904
+      "required_token_positions_per_second": 28813.30115915493,
+      "required_pages_per_second_upper_estimate": 0.3695860899572213
     }
   ],
   "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": 0.42295985438289563,
-    "producer_p95_seconds": 0.22495917700143764,
-    "gpu_update_seconds": 0.9538834405086157,
-    "producer_batches_per_second": 2.364290581334283,
-    "gpu_batches_per_second": 1.0483461160273362,
-    "producer_headroom": 2.255257634085261,
+    "producer_batch_seconds": 0.1519998379062315,
+    "producer_p95_seconds": 0.21432510100203217,
+    "gpu_update_seconds": 0.9595568327031255,
+    "producer_batches_per_second": 6.578954384259927,
+    "gpu_batches_per_second": 1.0421477560458237,
+    "producer_headroom": 6.312880631458797,
     "sustainable": true,
-    "buffer_seconds": 60.09465675204279,
+    "buffer_seconds": 60.452080460296905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387413.60983117303,
+    "generation_bytes_per_second": 392167.2770069212,
     "upload_bytes_per_second": 3680890.701886359,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.12994214249986,
+    "estimated_full_shards_per_hour": 84.14996845870711,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 941800, "loss": 0.5754478573799133, "lr": 3e-05, "grad_norm": 0.34232803215135954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137852720991, "tokens_per_second": 153008.36412336267, "data_wait_s": 1.0665000445442274e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941900, "loss": 0.49831411242485046, "lr": 3e-05, "grad_norm": 0.3497331215042466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137867413127, "tokens_per_second": 152633.72317608906, "data_wait_s": 1.2627999240066856e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942000, "loss": 0.5415322184562683, "lr": 3e-05, "grad_norm": 0.3412821098691674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137881971065, "tokens_per_second": 151928.18644188144, "data_wait_s": 1.5634002920705825e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 942100, "loss": 0.5240259766578674, "lr": 3e-05, "grad_norm": 0.3313426137066382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137896578923, "tokens_per_second": 151914.21895016293, "data_wait_s": 1.0146999557036906e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5b7f7465918a192e542ffcb27d5f781ec7ddda602253f847b86bf6bb80b2f72
-size 362574
+oid sha256:a950236f3597cd30969126469dd75e587de6c9895164e57b18e849125c70e712
+size 364692