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": 76197.3567430945,
-      "required_pages_per_second_upper_estimate": 0.005271509941713514
+      "required_token_positions_per_second": 76594.44476146046,
+      "required_pages_per_second_upper_estimate": 0.005298981412195214
     },
     {
       "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": 9524.669592886812,
-      "required_pages_per_second_upper_estimate": 0.003987780260823195
+      "required_token_positions_per_second": 9574.305595182557,
+      "required_pages_per_second_upper_estimate": 0.00400856181846683
     },
     {
       "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": 19049.339185773624,
-      "required_pages_per_second_upper_estimate": 0.003967474987539744
+      "required_token_positions_per_second": 19148.611190365114,
+      "required_pages_per_second_upper_estimate": 0.003988150728117338
     },
     {
       "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": 28574.008778660434,
-      "required_pages_per_second_upper_estimate": 0.001712968196295546
+      "required_token_positions_per_second": 28722.916785547673,
+      "required_pages_per_second_upper_estimate": 0.0017218950039390747
     },
     {
       "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": 28574.008778660434,
-      "required_pages_per_second_upper_estimate": 0.002221503498758903
+      "required_token_positions_per_second": 28722.916785547673,
+      "required_pages_per_second_upper_estimate": 0.0022330804413172836
     },
     {
       "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": 28574.008778660434,
-      "required_pages_per_second_upper_estimate": 0.36651670423237814
+      "required_token_positions_per_second": 28722.916785547673,
+      "required_pages_per_second_upper_estimate": 0.36842673625976674
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.26680382281233506,
-    "producer_p95_seconds": 0.3381283380003879,
-    "gpu_update_seconds": 0.9675926193684102,
-    "producer_batches_per_second": 3.7480722332205176,
-    "gpu_batches_per_second": 1.0334927943670587,
-    "producer_headroom": 3.6266070297238477,
+    "producer_batch_seconds": 0.13781230119496968,
+    "producer_p95_seconds": 0.3327732869947795,
+    "gpu_update_seconds": 0.9625763360464654,
+    "producer_batches_per_second": 7.256246295352488,
+    "gpu_batches_per_second": 1.0388786453106074,
+    "producer_headroom": 6.984690972431136,
     "sustainable": true,
-    "buffer_seconds": 60.958335020209844,
+    "buffer_seconds": 60.64230917092732,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436323.1897342358,
-    "upload_bytes_per_second": 2894102.4557824824,
+    "generation_bytes_per_second": 433973.6864316839,
+    "upload_bytes_per_second": 2903125.768790339,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.62479943294817,
+    "estimated_full_shards_per_hour": 93.1206507178582,
     "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": 2.9534675750037422,
-      "bytes": 5286995,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.3232240569996065,
       "bytes": 8048155,
@@ -562,6 +556,12 @@
       "bytes": 16202816,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.6036837479987298,
+      "bytes": 2721542,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -432,3 +432,4 @@
 {"step": 885700, "loss": 0.5637556910514832, "lr": 3e-05, "grad_norm": 0.3169692975692317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129641887771, "tokens_per_second": 153566.3688911708, "data_wait_s": 1.7531005141790956e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885800, "loss": 0.5134627223014832, "lr": 3e-05, "grad_norm": 0.3256320482823105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129656413302, "tokens_per_second": 150971.0908989782, "data_wait_s": 1.4251003449317068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 885900, "loss": 0.6447805762290955, "lr": 3e-05, "grad_norm": 0.35156226701477516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129670955519, "tokens_per_second": 150993.02279477424, "data_wait_s": 1.1322001228109002e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 886000, "loss": 0.54771488904953, "lr": 3e-05, "grad_norm": 0.31169686749996584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129685560151, "tokens_per_second": 151801.6015801023, "data_wait_s": 1.4566001482307911e-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 sha256:1bc94569984e11cae46cbe90ee42f64991e767b2818b82182d3f0d2538b307b4
-size 926424
+oid sha256:7d367a58427cdb908e6323c812197fa161ed6bee76bcbe29c234fdd3d9c2e916
+size 928542