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": 76780.02194900489,
-      "required_pages_per_second_upper_estimate": 0.0053118200726279385
+      "required_token_positions_per_second": 74986.12666770034,
+      "required_pages_per_second_upper_estimate": 0.005187714234656769
     },
     {
       "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": 9597.502743625611,
-      "required_pages_per_second_upper_estimate": 0.004018273980108392
+      "required_token_positions_per_second": 9373.265833462543,
+      "required_pages_per_second_upper_estimate": 0.003924390668422276
     },
     {
       "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": 19195.005487251223,
-      "required_pages_per_second_upper_estimate": 0.003997813436658829
+      "required_token_positions_per_second": 18746.531666925086,
+      "required_pages_per_second_upper_estimate": 0.0039044081669348716
     },
     {
       "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": 28792.508230876836,
-      "required_pages_per_second_upper_estimate": 0.0017260669048265735
+      "required_token_positions_per_second": 28119.79750038763,
+      "required_pages_per_second_upper_estimate": 0.0016857389237035561
     },
     {
       "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": 28792.508230876836,
-      "required_pages_per_second_upper_estimate": 0.0022384908700912075
+      "required_token_positions_per_second": 28119.79750038763,
+      "required_pages_per_second_upper_estimate": 0.002186190569737465
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28792.508230876836,
-      "required_pages_per_second_upper_estimate": 0.3693193805989769
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28119.79750038763,
+      "required_pages_per_second_upper_estimate": 0.3606905696487684
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11951544600009356,
-    "producer_p95_seconds": 0.36750896599551197,
-    "gpu_update_seconds": 0.9602497906156896,
-    "producer_batches_per_second": 8.367119342877382,
-    "gpu_batches_per_second": 1.0413956970079874,
-    "producer_headroom": 8.034524597054492,
+    "producer_batch_seconds": 0.4376694641797485,
+    "producer_p95_seconds": 0.35923797600116814,
+    "gpu_update_seconds": 0.9832218741837978,
+    "producer_batches_per_second": 2.2848292646464032,
+    "gpu_batches_per_second": 1.0170644350545295,
+    "producer_headroom": 2.246494111775625,
     "sustainable": true,
-    "buffer_seconds": 60.49573680878844,
+    "buffer_seconds": 60.959756199395464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346636.13747842767,
-    "upload_bytes_per_second": 4181553.2178794052,
+    "generation_bytes_per_second": 300487.84409752546,
+    "upload_bytes_per_second": 4153695.320426267,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.38004582657454,
+    "estimated_full_shards_per_hour": 64.47769634432147,
     "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": 1.4420257410019985,
-      "bytes": 5442386,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 2.6930293979967246,
       "bytes": 5991867,
@@ -562,6 +556,12 @@
       "bytes": 7832230,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 0.9895193189950078,
+      "bytes": 741196,
+      "pages": 23,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -402,3 +402,4 @@
 {"step": 840200, "loss": 0.5049573183059692, "lr": 3e-05, "grad_norm": 0.3118878406648667, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122981334289, "tokens_per_second": 151254.18466818944, "data_wait_s": 1.0846997611224651e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840300, "loss": 0.5117914080619812, "lr": 3e-05, "grad_norm": 0.2940921969676939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122995970860, "tokens_per_second": 151514.29915577834, "data_wait_s": 8.223003533203155e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 840400, "loss": 0.538476288318634, "lr": 3e-05, "grad_norm": 0.2918500214189647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123010790877, "tokens_per_second": 152886.55978331104, "data_wait_s": 1.164599962066859e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 840500, "loss": 0.5224630236625671, "lr": 3e-05, "grad_norm": 0.29421269579808434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123025473590, "tokens_per_second": 152329.25091578267, "data_wait_s": 1.0652998753357679e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2af1910829298f96f318c3e66de1b9612800d865915b6f6bc6a4486c5e4509a
-size 862101
+oid sha256:2952a2faaae768d8f82de4f6839976b7c8a61985ce904385b1c19b28e9eff5f1
+size 864219