Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75464.82589855166,
-      "required_pages_per_second_upper_estimate": 0.006541099845033837
+      "required_token_positions_per_second": 75141.4854080365,
+      "required_pages_per_second_upper_estimate": 0.006513073510814964
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9433.103237318957,
-      "required_pages_per_second_upper_estimate": 0.0037630059188283695
+      "required_token_positions_per_second": 9392.685676004563,
+      "required_pages_per_second_upper_estimate": 0.0037468827493236647
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18866.206474637915,
-      "required_pages_per_second_upper_estimate": 0.0043643648340138935
+      "required_token_positions_per_second": 18785.371352009126,
+      "required_pages_per_second_upper_estimate": 0.004345665051043292
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28299.309711956874,
-      "required_pages_per_second_upper_estimate": 0.0017232794864548678
+      "required_token_positions_per_second": 28178.05702801369,
+      "required_pages_per_second_upper_estimate": 0.0017158958341663953
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28299.309711956874,
-      "required_pages_per_second_upper_estimate": 0.0021785591991295462
+      "required_token_positions_per_second": 28178.05702801369,
+      "required_pages_per_second_upper_estimate": 0.002169224831870688
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28299.309711956874,
-      "required_pages_per_second_upper_estimate": 0.35138271492552337
+      "required_token_positions_per_second": 28178.05702801369,
+      "required_pages_per_second_upper_estimate": 0.3498771623975774
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08258479545310138,
-    "producer_p95_seconds": 0.20058339500064903,
-    "gpu_update_seconds": 0.9769849611673854,
-    "producer_batches_per_second": 12.108766444398164,
-    "gpu_batches_per_second": 1.0235572089104772,
-    "producer_headroom": 11.830082714465279,
+    "producer_batch_seconds": 0.10589830667971967,
+    "producer_p95_seconds": 0.22913751899977797,
+    "gpu_update_seconds": 0.9811890142926916,
+    "producer_batches_per_second": 9.443021624740556,
+    "gpu_batches_per_second": 1.0191716228303562,
+    "producer_headroom": 9.265389079923755,
     "sustainable": true,
-    "buffer_seconds": 60.573067592377896,
+    "buffer_seconds": 60.833718886146876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 234105.35926106802,
-    "upload_bytes_per_second": 5861930.493580846,
+    "generation_bytes_per_second": 234153.94350545335,
+    "upload_bytes_per_second": 5801164.826842243,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.233560403576185,
+    "estimated_full_shards_per_hour": 50.24398545143796,
     "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": 1.0847937970002022,
-      "bytes": 7859347,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2897943269999814,
       "bytes": 16445630,
@@ -562,6 +556,12 @@
       "bytes": 9539459,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.8906719839997095,
+      "bytes": 2274847,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 367900, "loss": 0.5798914544284344, "lr": 0.0003, "grad_norm": 0.17950501109424952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53850531284, "tokens_per_second": 146687.46292830593, "data_wait_s": 1.8110000382876024e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368000, "loss": 0.6261022314429283, "lr": 0.0003, "grad_norm": 0.14413710190247747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53865083427, "tokens_per_second": 148231.37495549192, "data_wait_s": 1.678199987509288e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368100, "loss": 0.5709961391985416, "lr": 0.0003, "grad_norm": 0.15939776561797428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53879659527, "tokens_per_second": 148466.76451177162, "data_wait_s": 1.598699964233674e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 368200, "loss": 0.5461985785514116, "lr": 0.0003, "grad_norm": 0.16204623314239167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53894327568, "tokens_per_second": 149660.57647001426, "data_wait_s": 1.8304999684914947e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f44bc00bb035acb5a009ad1cba5fa54c83e08acb853dc9595a182866b4c7c47
-size 279189
+oid sha256:6f9f32c3c8ec7b8881caadd49e0f513d4481950ce06eabf7d487d957f1fa38cb
+size 281307