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": 77302.13115816207,
-      "required_pages_per_second_upper_estimate": 0.005347940799177704
+      "required_token_positions_per_second": 77334.33728593063,
+      "required_pages_per_second_upper_estimate": 0.0053501688938252445
     },
     {
       "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": 9662.766394770259,
-      "required_pages_per_second_upper_estimate": 0.004045598507982645
+      "required_token_positions_per_second": 9666.792160741328,
+      "required_pages_per_second_upper_estimate": 0.004047284012127178
     },
     {
       "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": 19325.532789540517,
-      "required_pages_per_second_upper_estimate": 0.004024998831489247
+      "required_token_positions_per_second": 19333.584321482656,
+      "required_pages_per_second_upper_estimate": 0.004026675753259619
     },
     {
       "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": 28988.299184310774,
-      "required_pages_per_second_upper_estimate": 0.0017378042735294624
+      "required_token_positions_per_second": 29000.376482223986,
+      "required_pages_per_second_upper_estimate": 0.0017385282890983933
     },
     {
       "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": 28988.299184310774,
-      "required_pages_per_second_upper_estimate": 0.002253712755527306
+      "required_token_positions_per_second": 29000.376482223986,
+      "required_pages_per_second_upper_estimate": 0.0022546517123176388
     },
     {
       "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": 28988.299184310774,
-      "required_pages_per_second_upper_estimate": 0.37183077672568043
+      "required_token_positions_per_second": 29000.376482223986,
+      "required_pages_per_second_upper_estimate": 0.37198569133571896
     }
   ],
   "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": 0.13289876807021983,
-    "producer_p95_seconds": 0.31691697299902444,
-    "gpu_update_seconds": 0.9537641316660558,
-    "producer_batches_per_second": 7.524524226376795,
-    "gpu_batches_per_second": 1.048477256376981,
-    "producer_headroom": 7.176621314970464,
+    "producer_batch_seconds": 0.08565420021847103,
+    "producer_p95_seconds": 0.3067962390014145,
+    "gpu_update_seconds": 0.9533669335964333,
+    "producer_batches_per_second": 11.674850707255258,
+    "gpu_batches_per_second": 1.0489140799415504,
+    "producer_headroom": 11.130416618972095,
     "sustainable": true,
-    "buffer_seconds": 60.087140294961515,
+    "buffer_seconds": 60.0621168165753,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456973.25008880463,
-    "upload_bytes_per_second": 5014547.921576481,
+    "generation_bytes_per_second": 457042.83396935405,
+    "upload_bytes_per_second": 4962740.959041732,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.05582167623619,
+    "estimated_full_shards_per_hour": 98.07075275717226,
     "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": 1.1610033780016238,
-      "bytes": 10490094,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.0719913379980426,
       "bytes": 2751265,
@@ -562,6 +556,12 @@
       "bytes": 1188999,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.4343414930008294,
+      "bytes": 7437168,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 589600, "loss": 0.5750231146812439, "lr": 0.0003, "grad_norm": 0.20048132500355573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86299595829, "tokens_per_second": 150362.5483587851, "data_wait_s": 1.089699799194932e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589700, "loss": 0.5951194763183594, "lr": 0.0003, "grad_norm": 0.153565967132763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86314303566, "tokens_per_second": 152970.13031007777, "data_wait_s": 1.2247000995557755e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 589800, "loss": 0.5887827277183533, "lr": 0.0003, "grad_norm": 0.18166399748246206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86329036117, "tokens_per_second": 153782.57758772705, "data_wait_s": 1.3060998753644526e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 589900, "loss": 0.6164164543151855, "lr": 0.0003, "grad_norm": 0.18821315083376589, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86343593442, "tokens_per_second": 150996.3781917091, "data_wait_s": 1.7334001313429326e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61ca3cdf68396f4da8d2c3145d765ed4f3bcbd530b565634ee573cbd0099fefb
-size 697449
+oid sha256:68815f3e25bf3a8eedefafb5acf011d38946336e0556d32e727365e0ec4b34fd
+size 699567