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": 77544.2492190025,
-      "required_pages_per_second_upper_estimate": 0.005364691088417945
+      "required_token_positions_per_second": 77570.23913567029,
+      "required_pages_per_second_upper_estimate": 0.005366489131157938
     },
     {
       "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": 9693.031152375312,
-      "required_pages_per_second_upper_estimate": 0.004058269729991874
+      "required_token_positions_per_second": 9696.279891958786,
+      "required_pages_per_second_upper_estimate": 0.004059629909414077
     },
     {
       "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": 19386.062304750623,
-      "required_pages_per_second_upper_estimate": 0.004037605533236852
+      "required_token_positions_per_second": 19392.55978391757,
+      "required_pages_per_second_upper_estimate": 0.004038958786797279
     },
     {
       "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": 29079.093457125935,
-      "required_pages_per_second_upper_estimate": 0.001743247251549904
+      "required_token_positions_per_second": 29088.839675876356,
+      "required_pages_per_second_upper_estimate": 0.0017438315224823772
     },
     {
       "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": 29079.093457125935,
-      "required_pages_per_second_upper_estimate": 0.0022607716108768846
+      "required_token_positions_per_second": 29088.839675876356,
+      "required_pages_per_second_upper_estimate": 0.002261529336515644
     },
     {
       "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": 29079.093457125935,
-      "required_pages_per_second_upper_estimate": 0.37299538817005856
+      "required_token_positions_per_second": 29088.839675876356,
+      "required_pages_per_second_upper_estimate": 0.3731204021995146
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10933117565625139,
-    "producer_p95_seconds": 0.18552211200039892,
-    "gpu_update_seconds": 0.9507861736048983,
-    "producer_batches_per_second": 9.14652196866614,
-    "gpu_batches_per_second": 1.0517611927490573,
-    "producer_headroom": 8.696386624381221,
+    "producer_batch_seconds": 0.4366551351327246,
+    "producer_p95_seconds": 0.20328514599896153,
+    "gpu_update_seconds": 0.9504676125988188,
+    "producer_batches_per_second": 2.2901368140238234,
+    "gpu_batches_per_second": 1.0521137035545558,
+    "producer_headroom": 2.176700870149889,
     "sustainable": true,
-    "buffer_seconds": 60.85031511071349,
+    "buffer_seconds": 60.829927206324406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326892.39159453986,
-    "upload_bytes_per_second": 4475677.988999582,
+    "generation_bytes_per_second": 282233.0441739166,
+    "upload_bytes_per_second": 4555266.0227301875,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.14349757077358,
+    "estimated_full_shards_per_hour": 60.56064123070835,
     "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": 0.8045400529999824,
-      "bytes": 1220702,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.144875766000041,
       "bytes": 9911722,
@@ -562,6 +556,12 @@
       "bytes": 2681477,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0707301010006631,
+      "bytes": 10039688,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -91,3 +91,4 @@
 {"step": 449300, "loss": 0.5931718349456787, "lr": 0.0003, "grad_norm": 0.14727832786485578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65763482923, "tokens_per_second": 152078.53892100567, "data_wait_s": 1.4055998690309934e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449400, "loss": 0.5917835831642151, "lr": 0.0003, "grad_norm": 0.16826501700247157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65778230056, "tokens_per_second": 153351.57652372826, "data_wait_s": 1.408600110153202e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 449500, "loss": 0.6102327704429626, "lr": 0.0003, "grad_norm": 0.14200678050190313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65792922550, "tokens_per_second": 153365.92923557613, "data_wait_s": 9.314999260823242e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 449600, "loss": 0.6116858720779419, "lr": 0.0003, "grad_norm": 0.15781677754937184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65807490947, "tokens_per_second": 122341.90125808115, "data_wait_s": 1.4140001439955086e-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 sha256:244b4c5a0ba78877b84337bea4a2e753cb44c49a9a0dbca2e9d065a2ad8ad8c4
-size 197922
+oid sha256:f2964a7ed7710c64d092e40c408933e6013f2ddbb1c03461cec33db088f85089
+size 200040